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 page_cache_inconsistent
:1;
161 bool needs_alignment
;
163 bool check_cache_dropped
;
168 typedef struct BDRVRawReopenState
{
172 bool check_cache_dropped
;
173 } BDRVRawReopenState
;
175 static int fd_open(BlockDriverState
*bs
);
176 static int64_t raw_getlength(BlockDriverState
*bs
);
178 typedef struct RawPosixAIOData
{
179 BlockDriverState
*bs
;
200 PreallocMode prealloc
;
206 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
207 static int cdrom_reopen(BlockDriverState
*bs
);
210 #if defined(__NetBSD__)
211 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
213 static char namebuf
[PATH_MAX
];
214 const char *dp
, *fname
;
218 dp
= strrchr(fname
, '/');
219 if (lstat(fname
, &sb
) < 0) {
220 error_setg_file_open(errp
, errno
, fname
);
224 if (!S_ISBLK(sb
.st_mode
)) {
229 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
231 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
232 (int)(dp
- fname
), fname
, dp
+ 1);
235 warn_report("%s is a block device, using %s", fname
, *filename
);
240 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
247 * Get logical block size via ioctl. On success store it in @sector_size_p.
249 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
251 unsigned int sector_size
;
252 bool success
= false;
256 static const unsigned long ioctl_list
[] = {
260 #ifdef DKIOCGETBLOCKSIZE
263 #ifdef DIOCGSECTORSIZE
268 /* Try a few ioctls to get the right size */
269 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
270 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
271 *sector_size_p
= sector_size
;
276 return success
? 0 : -errno
;
280 * Get physical block size of @fd.
281 * On success, store it in @blk_size and return 0.
282 * On failure, return -errno.
284 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
287 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
296 /* Check if read is allowed with given memory buffer and length.
298 * This function is used to check O_DIRECT memory buffer and request alignment.
300 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
302 ssize_t ret
= pread(fd
, buf
, len
, 0);
309 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
310 * other errors (e.g. real I/O error), which could happen on a failed
311 * drive, since we only care about probing alignment.
313 if (errno
!= EINVAL
) {
321 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
323 BDRVRawState
*s
= bs
->opaque
;
325 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
326 size_t alignments
[] = {1, 512, 1024, 2048, 4096};
328 /* For SCSI generic devices the alignment is not really used.
329 With buffered I/O, we don't have any restrictions. */
330 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
331 bs
->bl
.request_alignment
= 1;
336 bs
->bl
.request_alignment
= 0;
338 /* Let's try to use the logical blocksize for the alignment. */
339 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
340 bs
->bl
.request_alignment
= 0;
345 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
346 bs
->bl
.request_alignment
= da
.d_miniosz
;
347 /* The kernel returns wrong information for d_mem */
348 /* s->buf_align = da.d_mem; */
354 * If we could not get the sizes so far, we can only guess them. First try
355 * to detect request alignment, since it is more likely to succeed. Then
356 * try to detect buf_align, which cannot be detected in some cases (e.g.
357 * Gluster). If buf_align cannot be detected, we fallback to the value of
361 if (!bs
->bl
.request_alignment
) {
364 buf
= qemu_memalign(max_align
, max_align
);
365 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
366 align
= alignments
[i
];
367 if (raw_is_io_aligned(fd
, buf
, align
)) {
368 /* Fallback to safe value. */
369 bs
->bl
.request_alignment
= (align
!= 1) ? align
: max_align
;
379 buf
= qemu_memalign(max_align
, 2 * max_align
);
380 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
381 align
= alignments
[i
];
382 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
383 /* Fallback to request_alignment. */
384 s
->buf_align
= (align
!= 1) ? align
: bs
->bl
.request_alignment
;
391 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
392 error_setg(errp
, "Could not find working O_DIRECT alignment");
393 error_append_hint(errp
, "Try cache.direct=off\n");
397 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
399 bool read_write
= false;
400 assert(open_flags
!= NULL
);
402 *open_flags
|= O_BINARY
;
403 *open_flags
&= ~O_ACCMODE
;
405 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
406 read_write
= has_writers
;
407 } else if (bdrv_flags
& BDRV_O_RDWR
) {
412 *open_flags
|= O_RDWR
;
414 *open_flags
|= O_RDONLY
;
417 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
418 * and O_DIRECT for no caching. */
419 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
420 *open_flags
|= O_DIRECT
;
424 static void raw_parse_filename(const char *filename
, QDict
*options
,
427 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
430 static QemuOptsList raw_runtime_opts
= {
432 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
436 .type
= QEMU_OPT_STRING
,
437 .help
= "File name of the image",
441 .type
= QEMU_OPT_STRING
,
442 .help
= "host AIO implementation (threads, native)",
446 .type
= QEMU_OPT_STRING
,
447 .help
= "file locking mode (on/off/auto, default: auto)",
450 .name
= "pr-manager",
451 .type
= QEMU_OPT_STRING
,
452 .help
= "id of persistent reservation manager object (default: none)",
454 #if defined(__linux__)
456 .name
= "drop-cache",
457 .type
= QEMU_OPT_BOOL
,
458 .help
= "invalidate page cache during live migration (default: on)",
462 .name
= "x-check-cache-dropped",
463 .type
= QEMU_OPT_BOOL
,
464 .help
= "check that page cache was dropped on live migration (default: off)"
466 { /* end of list */ }
470 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
472 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
473 int bdrv_flags
, int open_flags
,
474 bool device
, Error
**errp
)
476 BDRVRawState
*s
= bs
->opaque
;
478 Error
*local_err
= NULL
;
479 const char *filename
= NULL
;
481 BlockdevAioOptions aio
, aio_default
;
486 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
487 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
489 error_propagate(errp
, local_err
);
494 filename
= qemu_opt_get(opts
, "filename");
496 ret
= raw_normalize_devicepath(&filename
, errp
);
501 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
502 ? BLOCKDEV_AIO_OPTIONS_NATIVE
503 : BLOCKDEV_AIO_OPTIONS_THREADS
;
504 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
505 qemu_opt_get(opts
, "aio"),
506 aio_default
, &local_err
);
508 error_propagate(errp
, local_err
);
512 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
514 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
515 qemu_opt_get(opts
, "locking"),
516 ON_OFF_AUTO_AUTO
, &local_err
);
518 error_propagate(errp
, local_err
);
525 if (!qemu_has_ofd_lock()) {
526 warn_report("File lock requested but OFD locking syscall is "
527 "unavailable, falling back to POSIX file locks");
528 error_printf("Due to the implementation, locks can be lost "
532 case ON_OFF_AUTO_OFF
:
535 case ON_OFF_AUTO_AUTO
:
536 s
->use_lock
= qemu_has_ofd_lock();
542 str
= qemu_opt_get(opts
, "pr-manager");
544 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
546 error_propagate(errp
, local_err
);
552 s
->drop_cache
= qemu_opt_get_bool(opts
, "drop-cache", true);
553 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
556 s
->open_flags
= open_flags
;
557 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
560 fd
= qemu_open(filename
, s
->open_flags
, 0644);
561 ret
= fd
< 0 ? -errno
: 0;
564 error_setg_file_open(errp
, -ret
, filename
);
573 s
->shared_perm
= BLK_PERM_ALL
;
575 #ifdef CONFIG_LINUX_AIO
576 /* Currently Linux does AIO only for files opened with O_DIRECT */
577 if (s
->use_linux_aio
) {
578 if (!(s
->open_flags
& O_DIRECT
)) {
579 error_setg(errp
, "aio=native was specified, but it requires "
580 "cache.direct=on, which was not specified.");
584 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
585 error_prepend(errp
, "Unable to use native AIO: ");
590 if (s
->use_linux_aio
) {
591 error_setg(errp
, "aio=native was specified, but is not supported "
596 #endif /* !defined(CONFIG_LINUX_AIO) */
598 s
->has_discard
= true;
599 s
->has_write_zeroes
= true;
600 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
601 s
->needs_alignment
= true;
604 if (fstat(s
->fd
, &st
) < 0) {
606 error_setg_errno(errp
, errno
, "Could not stat file");
611 if (S_ISBLK(st
.st_mode
)) {
612 warn_report("Opening a block device as a file using the '%s' "
613 "driver is deprecated", bs
->drv
->format_name
);
614 } else if (S_ISCHR(st
.st_mode
)) {
615 warn_report("Opening a character device as a file using the '%s' "
616 "driver is deprecated", bs
->drv
->format_name
);
617 } else if (!S_ISREG(st
.st_mode
)) {
618 error_setg(errp
, "A regular file was expected by the '%s' driver, "
619 "but something else was given", bs
->drv
->format_name
);
623 s
->discard_zeroes
= true;
624 s
->has_fallocate
= true;
627 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
628 error_setg(errp
, "'%s' driver expects either "
629 "a character or block device", bs
->drv
->format_name
);
635 if (S_ISBLK(st
.st_mode
)) {
636 #ifdef BLKDISCARDZEROES
638 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
639 s
->discard_zeroes
= true;
643 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
644 * not rely on the contents of discarded blocks unless using O_DIRECT.
645 * Same for BLKZEROOUT.
647 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
648 s
->discard_zeroes
= false;
649 s
->has_write_zeroes
= false;
654 if (S_ISCHR(st
.st_mode
)) {
656 * The file is a char device (disk), which on FreeBSD isn't behind
657 * a pager, so force all requests to be aligned. This is needed
658 * so QEMU makes sure all IO operations on the device are aligned
659 * to sector size, or else FreeBSD will reject them with EINVAL.
661 s
->needs_alignment
= true;
666 if (platform_test_xfs_fd(s
->fd
)) {
671 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
674 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
681 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
684 BDRVRawState
*s
= bs
->opaque
;
686 s
->type
= FTYPE_FILE
;
687 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
696 #define PERM_FOREACH(i) \
697 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
699 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
700 * file; if @unlock == true, also unlock the unneeded bytes.
701 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
703 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
704 uint64_t perm_lock_bits
,
705 uint64_t shared_perm_lock_bits
,
706 bool unlock
, Error
**errp
)
710 uint64_t locked_perm
, locked_shared_perm
;
713 locked_perm
= s
->locked_perm
;
714 locked_shared_perm
= s
->locked_shared_perm
;
717 * We don't have the previous bits, just lock/unlock for each of the
721 locked_perm
= BLK_PERM_ALL
;
722 locked_shared_perm
= BLK_PERM_ALL
;
725 locked_shared_perm
= 0;
730 int off
= RAW_LOCK_PERM_BASE
+ i
;
731 uint64_t bit
= (1ULL << i
);
732 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
733 ret
= qemu_lock_fd(fd
, off
, 1, false);
735 error_setg(errp
, "Failed to lock byte %d", off
);
738 s
->locked_perm
|= bit
;
740 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
741 ret
= qemu_unlock_fd(fd
, off
, 1);
743 error_setg(errp
, "Failed to unlock byte %d", off
);
746 s
->locked_perm
&= ~bit
;
751 int off
= RAW_LOCK_SHARED_BASE
+ i
;
752 uint64_t bit
= (1ULL << i
);
753 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
754 ret
= qemu_lock_fd(fd
, off
, 1, false);
756 error_setg(errp
, "Failed to lock byte %d", off
);
759 s
->locked_shared_perm
|= bit
;
761 } else if (unlock
&& (locked_shared_perm
& bit
) &&
762 !(shared_perm_lock_bits
& bit
)) {
763 ret
= qemu_unlock_fd(fd
, off
, 1);
765 error_setg(errp
, "Failed to unlock byte %d", off
);
768 s
->locked_shared_perm
&= ~bit
;
775 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
776 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
783 int off
= RAW_LOCK_SHARED_BASE
+ i
;
784 uint64_t p
= 1ULL << i
;
786 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
788 char *perm_name
= bdrv_perm_names(p
);
790 "Failed to get \"%s\" lock",
798 int off
= RAW_LOCK_PERM_BASE
+ i
;
799 uint64_t p
= 1ULL << i
;
800 if (!(shared_perm
& p
)) {
801 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
803 char *perm_name
= bdrv_perm_names(p
);
805 "Failed to get shared \"%s\" lock",
815 static int raw_handle_perm_lock(BlockDriverState
*bs
,
817 uint64_t new_perm
, uint64_t new_shared
,
820 BDRVRawState
*s
= bs
->opaque
;
822 Error
*local_err
= NULL
;
828 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
834 if ((s
->perm
| new_perm
) == s
->perm
&&
835 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
838 * We are going to unlock bytes, it should not fail. If it fail due
839 * to some fs-dependent permission-unrelated reasons (which occurs
840 * sometimes on NFS and leads to abort in bdrv_replace_child) we
841 * can't prevent such errors by any check here. And we ignore them
842 * anyway in ABORT and COMMIT.
846 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
847 ~s
->shared_perm
| ~new_shared
,
850 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
854 error_append_hint(errp
,
855 "Is another process using the image [%s]?\n",
859 /* fall through to unlock bytes. */
861 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
864 /* Theoretically the above call only unlocks bytes and it cannot
865 * fail. Something weird happened, report it.
867 warn_report_err(local_err
);
871 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
874 /* Theoretically the above call only unlocks bytes and it cannot
875 * fail. Something weird happened, report it.
877 warn_report_err(local_err
);
884 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
885 int *open_flags
, uint64_t perm
, bool force_dup
,
888 BDRVRawState
*s
= bs
->opaque
;
891 bool has_writers
= perm
&
892 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
893 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
895 fcntl_flags
|= O_NOATIME
;
899 if (s
->type
== FTYPE_CD
) {
900 *open_flags
|= O_NONBLOCK
;
903 raw_parse_flags(flags
, open_flags
, has_writers
);
906 /* Not all operating systems have O_ASYNC, and those that don't
907 * will not let us track the state into rs->open_flags (typically
908 * you achieve the same effect with an ioctl, for example I_SETSIG
909 * on Solaris). But we do not use O_ASYNC, so that's fine.
911 assert((s
->open_flags
& O_ASYNC
) == 0);
914 if (!force_dup
&& *open_flags
== s
->open_flags
) {
915 /* We're lucky, the existing fd is fine */
919 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
920 /* dup the original fd */
921 fd
= qemu_dup(s
->fd
);
923 ret
= fcntl_setfl(fd
, *open_flags
);
931 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
933 const char *normalized_filename
= bs
->filename
;
934 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
936 assert(!(*open_flags
& O_CREAT
));
937 fd
= qemu_open(normalized_filename
, *open_flags
);
939 error_setg_errno(errp
, errno
, "Could not reopen file");
948 static int raw_reopen_prepare(BDRVReopenState
*state
,
949 BlockReopenQueue
*queue
, Error
**errp
)
952 BDRVRawReopenState
*rs
;
955 Error
*local_err
= NULL
;
957 assert(state
!= NULL
);
958 assert(state
->bs
!= NULL
);
960 s
= state
->bs
->opaque
;
962 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
965 /* Handle options changes */
966 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
967 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
969 error_propagate(errp
, local_err
);
974 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
975 rs
->check_cache_dropped
=
976 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
978 /* This driver's reopen function doesn't currently allow changing
979 * other options, so let's put them back in the original QDict and
980 * bdrv_reopen_prepare() will detect changes and complain. */
981 qemu_opts_to_qdict(opts
, state
->options
);
983 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
984 state
->perm
, true, &local_err
);
986 error_propagate(errp
, local_err
);
991 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
992 * alignment with the new fd. */
994 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
996 error_propagate(errp
, local_err
);
1002 s
->reopen_state
= state
;
1010 qemu_opts_del(opts
);
1014 static void raw_reopen_commit(BDRVReopenState
*state
)
1016 BDRVRawReopenState
*rs
= state
->opaque
;
1017 BDRVRawState
*s
= state
->bs
->opaque
;
1019 s
->drop_cache
= rs
->drop_cache
;
1020 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1021 s
->open_flags
= rs
->open_flags
;
1026 g_free(state
->opaque
);
1027 state
->opaque
= NULL
;
1029 assert(s
->reopen_state
== state
);
1030 s
->reopen_state
= NULL
;
1034 static void raw_reopen_abort(BDRVReopenState
*state
)
1036 BDRVRawReopenState
*rs
= state
->opaque
;
1037 BDRVRawState
*s
= state
->bs
->opaque
;
1039 /* nothing to do if NULL, we didn't get far enough */
1048 g_free(state
->opaque
);
1049 state
->opaque
= NULL
;
1051 assert(s
->reopen_state
== state
);
1052 s
->reopen_state
= NULL
;
1055 static int sg_get_max_transfer_length(int fd
)
1060 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1070 static int sg_get_max_segments(int fd
)
1075 char *sysfspath
= NULL
;
1081 if (fstat(fd
, &st
)) {
1086 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1087 major(st
.st_rdev
), minor(st
.st_rdev
));
1088 sysfd
= open(sysfspath
, O_RDONLY
);
1094 ret
= read(sysfd
, buf
, sizeof(buf
) - 1);
1095 } while (ret
== -1 && errno
== EINTR
);
1099 } else if (ret
== 0) {
1104 /* The file is ended with '\n', pass 'end' to accept that. */
1105 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1106 if (ret
== 0 && end
&& *end
== '\n') {
1121 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1123 BDRVRawState
*s
= bs
->opaque
;
1126 int ret
= sg_get_max_transfer_length(s
->fd
);
1128 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1129 bs
->bl
.max_transfer
= pow2floor(ret
);
1132 ret
= sg_get_max_segments(s
->fd
);
1134 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
, ret
* getpagesize());
1138 raw_probe_alignment(bs
, s
->fd
, errp
);
1139 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1140 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1143 static int check_for_dasd(int fd
)
1146 struct dasd_information2_t info
= {0};
1148 return ioctl(fd
, BIODASDINFO2
, &info
);
1155 * Try to get @bs's logical and physical block size.
1156 * On success, store them in @bsz and return zero.
1157 * On failure, return negative errno.
1159 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1161 BDRVRawState
*s
= bs
->opaque
;
1164 /* If DASD, get blocksizes */
1165 if (check_for_dasd(s
->fd
) < 0) {
1168 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1172 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1176 * Try to get @bs's geometry: cyls, heads, sectors.
1177 * On success, store them in @geo and return 0.
1178 * On failure return -errno.
1179 * (Allows block driver to assign default geometry values that guest sees)
1182 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1184 BDRVRawState
*s
= bs
->opaque
;
1185 struct hd_geometry ioctl_geo
= {0};
1187 /* If DASD, get its geometry */
1188 if (check_for_dasd(s
->fd
) < 0) {
1191 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1194 /* HDIO_GETGEO may return success even though geo contains zeros
1195 (e.g. certain multipath setups) */
1196 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1199 /* Do not return a geometry for partition */
1200 if (ioctl_geo
.start
!= 0) {
1203 geo
->heads
= ioctl_geo
.heads
;
1204 geo
->sectors
= ioctl_geo
.sectors
;
1205 geo
->cylinders
= ioctl_geo
.cylinders
;
1209 #else /* __linux__ */
1210 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1216 #if defined(__linux__)
1217 static int handle_aiocb_ioctl(void *opaque
)
1219 RawPosixAIOData
*aiocb
= opaque
;
1222 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1231 static int handle_aiocb_flush(void *opaque
)
1233 RawPosixAIOData
*aiocb
= opaque
;
1234 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1237 if (s
->page_cache_inconsistent
) {
1241 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1243 /* There is no clear definition of the semantics of a failing fsync(),
1244 * so we may have to assume the worst. The sad truth is that this
1245 * assumption is correct for Linux. Some pages are now probably marked
1246 * clean in the page cache even though they are inconsistent with the
1247 * on-disk contents. The next fdatasync() call would succeed, but no
1248 * further writeback attempt will be made. We can't get back to a state
1249 * in which we know what is on disk (we would have to rewrite
1250 * everything that was touched since the last fdatasync() at least), so
1251 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1252 * really defined, I have little hope that other OSes are doing better.
1254 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1256 if ((s
->open_flags
& O_DIRECT
) == 0) {
1257 s
->page_cache_inconsistent
= true;
1264 #ifdef CONFIG_PREADV
1266 static bool preadv_present
= true;
1269 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1271 return preadv(fd
, iov
, nr_iov
, offset
);
1275 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1277 return pwritev(fd
, iov
, nr_iov
, offset
);
1282 static bool preadv_present
= false;
1285 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1291 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1298 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1303 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1304 len
= qemu_pwritev(aiocb
->aio_fildes
,
1309 len
= qemu_preadv(aiocb
->aio_fildes
,
1313 } while (len
== -1 && errno
== EINTR
);
1322 * Read/writes the data to/from a given linear buffer.
1324 * Returns the number of bytes handles or -errno in case of an error. Short
1325 * reads are only returned if the end of the file is reached.
1327 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1332 while (offset
< aiocb
->aio_nbytes
) {
1333 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1334 len
= pwrite(aiocb
->aio_fildes
,
1335 (const char *)buf
+ offset
,
1336 aiocb
->aio_nbytes
- offset
,
1337 aiocb
->aio_offset
+ offset
);
1339 len
= pread(aiocb
->aio_fildes
,
1341 aiocb
->aio_nbytes
- offset
,
1342 aiocb
->aio_offset
+ offset
);
1344 if (len
== -1 && errno
== EINTR
) {
1346 } else if (len
== -1 && errno
== EINVAL
&&
1347 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1348 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1350 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1351 * after a short read. Assume that O_DIRECT short reads only occur
1352 * at EOF. Therefore this is a short read, not an I/O error.
1355 } else if (len
== -1) {
1358 } else if (len
== 0) {
1367 static int handle_aiocb_rw(void *opaque
)
1369 RawPosixAIOData
*aiocb
= opaque
;
1373 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1375 * If there is just a single buffer, and it is properly aligned
1376 * we can just use plain pread/pwrite without any problems.
1378 if (aiocb
->io
.niov
== 1) {
1379 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1383 * We have more than one iovec, and all are properly aligned.
1385 * Try preadv/pwritev first and fall back to linearizing the
1386 * buffer if it's not supported.
1388 if (preadv_present
) {
1389 nbytes
= handle_aiocb_rw_vector(aiocb
);
1390 if (nbytes
== aiocb
->aio_nbytes
||
1391 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1394 preadv_present
= false;
1398 * XXX(hch): short read/write. no easy way to handle the reminder
1399 * using these interfaces. For now retry using plain
1405 * Ok, we have to do it the hard way, copy all segments into
1406 * a single aligned buffer.
1408 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1414 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1418 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1419 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1420 p
+= aiocb
->io
.iov
[i
].iov_len
;
1422 assert(p
- buf
== aiocb
->aio_nbytes
);
1425 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1426 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1428 size_t count
= aiocb
->aio_nbytes
, copy
;
1431 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1433 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1434 copy
= aiocb
->io
.iov
[i
].iov_len
;
1436 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1437 assert(count
>= copy
);
1446 if (nbytes
== aiocb
->aio_nbytes
) {
1448 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1449 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1452 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1453 0, aiocb
->aio_nbytes
- nbytes
);
1462 static int translate_err(int err
)
1464 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1471 #ifdef CONFIG_FALLOCATE
1472 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1475 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1478 } while (errno
== EINTR
);
1479 return translate_err(-errno
);
1483 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1486 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1488 if (!s
->has_write_zeroes
) {
1493 /* The BLKZEROOUT implementation in the kernel doesn't set
1494 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1496 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1498 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1499 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1502 } while (errno
== EINTR
);
1504 ret
= translate_err(-errno
);
1505 if (ret
== -ENOTSUP
) {
1506 s
->has_write_zeroes
= false;
1514 static int handle_aiocb_write_zeroes(void *opaque
)
1516 RawPosixAIOData
*aiocb
= opaque
;
1517 #ifdef CONFIG_FALLOCATE
1518 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1522 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1523 return handle_aiocb_write_zeroes_block(aiocb
);
1526 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1527 if (s
->has_write_zeroes
) {
1528 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1529 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1530 if (ret
== -EINVAL
) {
1532 * Allow falling back to pwrite for file systems that
1533 * do not support fallocate() for an unaligned byte range.
1537 if (ret
== 0 || ret
!= -ENOTSUP
) {
1540 s
->has_write_zeroes
= false;
1544 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1545 if (s
->has_discard
&& s
->has_fallocate
) {
1546 int ret
= do_fallocate(s
->fd
,
1547 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1548 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1550 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1551 if (ret
== 0 || ret
!= -ENOTSUP
) {
1554 s
->has_fallocate
= false;
1555 } else if (ret
!= -ENOTSUP
) {
1558 s
->has_discard
= false;
1563 #ifdef CONFIG_FALLOCATE
1564 /* Last resort: we are trying to extend the file with zeroed data. This
1565 * can be done via fallocate(fd, 0) */
1566 len
= bdrv_getlength(aiocb
->bs
);
1567 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1568 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1569 if (ret
== 0 || ret
!= -ENOTSUP
) {
1572 s
->has_fallocate
= false;
1579 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1581 RawPosixAIOData
*aiocb
= opaque
;
1582 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1585 /* First try to write zeros and unmap at the same time */
1587 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1588 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1589 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1590 if (ret
!= -ENOTSUP
) {
1595 /* If we couldn't manage to unmap while guaranteed that the area reads as
1596 * all-zero afterwards, just write zeroes without unmapping */
1597 ret
= handle_aiocb_write_zeroes(aiocb
);
1601 #ifndef HAVE_COPY_FILE_RANGE
1602 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1603 off_t
*out_off
, size_t len
, unsigned int flags
)
1605 #ifdef __NR_copy_file_range
1606 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1607 out_off
, len
, flags
);
1615 static int handle_aiocb_copy_range(void *opaque
)
1617 RawPosixAIOData
*aiocb
= opaque
;
1618 uint64_t bytes
= aiocb
->aio_nbytes
;
1619 off_t in_off
= aiocb
->aio_offset
;
1620 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1623 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1624 aiocb
->copy_range
.aio_fd2
, &out_off
,
1626 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1627 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1630 /* No progress (e.g. when beyond EOF), let the caller fall back to
1649 static int handle_aiocb_discard(void *opaque
)
1651 RawPosixAIOData
*aiocb
= opaque
;
1652 int ret
= -EOPNOTSUPP
;
1653 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1655 if (!s
->has_discard
) {
1659 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1662 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1663 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1666 } while (errno
== EINTR
);
1671 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1672 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1673 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1677 ret
= translate_err(ret
);
1678 if (ret
== -ENOTSUP
) {
1679 s
->has_discard
= false;
1685 * Help alignment probing by allocating the first block.
1687 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1688 * reading succeeds regardless of request length. In this case we fallback to
1689 * safe alignment which is not optimal. Allocating the first block avoids this
1692 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1693 * request alignment, so we use safe values.
1695 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1696 * caller may ignore failures.
1698 static int allocate_first_block(int fd
, size_t max_size
)
1700 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
1703 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
1708 buf
= qemu_memalign(max_align
, write_size
);
1709 memset(buf
, 0, write_size
);
1712 n
= pwrite(fd
, buf
, write_size
, 0);
1713 } while (n
== -1 && errno
== EINTR
);
1715 ret
= (n
== -1) ? -errno
: 0;
1721 static int handle_aiocb_truncate(void *opaque
)
1723 RawPosixAIOData
*aiocb
= opaque
;
1725 int64_t current_length
= 0;
1728 int fd
= aiocb
->aio_fildes
;
1729 int64_t offset
= aiocb
->aio_offset
;
1730 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1731 Error
**errp
= aiocb
->truncate
.errp
;
1733 if (fstat(fd
, &st
) < 0) {
1735 error_setg_errno(errp
, -result
, "Could not stat file");
1739 current_length
= st
.st_size
;
1740 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1741 error_setg(errp
, "Cannot use preallocation for shrinking files");
1746 #ifdef CONFIG_POSIX_FALLOCATE
1747 case PREALLOC_MODE_FALLOC
:
1749 * Truncating before posix_fallocate() makes it about twice slower on
1750 * file systems that do not support fallocate(), trying to check if a
1751 * block is allocated before allocating it, so don't do that here.
1753 if (offset
!= current_length
) {
1754 result
= -posix_fallocate(fd
, current_length
,
1755 offset
- current_length
);
1757 /* posix_fallocate() doesn't set errno. */
1758 error_setg_errno(errp
, -result
,
1759 "Could not preallocate new data");
1760 } else if (current_length
== 0) {
1762 * posix_fallocate() uses fallocate() if the filesystem
1763 * supports it, or fallback to manually writing zeroes. If
1764 * fallocate() was used, unaligned reads from the fallocated
1765 * area in raw_probe_alignment() will succeed, hence we need to
1766 * allocate the first block.
1768 * Optimize future alignment probing; ignore failures.
1770 allocate_first_block(fd
, offset
);
1777 case PREALLOC_MODE_FULL
:
1779 int64_t num
= 0, left
= offset
- current_length
;
1783 * Knowing the final size from the beginning could allow the file
1784 * system driver to do less allocations and possibly avoid
1785 * fragmentation of the file.
1787 if (ftruncate(fd
, offset
) != 0) {
1789 error_setg_errno(errp
, -result
, "Could not resize file");
1793 buf
= g_malloc0(65536);
1795 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1796 if (seek_result
< 0) {
1798 error_setg_errno(errp
, -result
,
1799 "Failed to seek to the old end of file");
1804 num
= MIN(left
, 65536);
1805 result
= write(fd
, buf
, num
);
1807 if (errno
== EINTR
) {
1811 error_setg_errno(errp
, -result
,
1812 "Could not write zeros for preallocation");
1821 error_setg_errno(errp
, -result
,
1822 "Could not flush file to disk");
1828 case PREALLOC_MODE_OFF
:
1829 if (ftruncate(fd
, offset
) != 0) {
1831 error_setg_errno(errp
, -result
, "Could not resize file");
1832 } else if (current_length
== 0 && offset
> current_length
) {
1833 /* Optimize future alignment probing; ignore failures. */
1834 allocate_first_block(fd
, offset
);
1839 error_setg(errp
, "Unsupported preallocation mode: %s",
1840 PreallocMode_str(prealloc
));
1846 if (ftruncate(fd
, current_length
) < 0) {
1847 error_report("Failed to restore old file length: %s",
1856 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1857 ThreadPoolFunc func
, void *arg
)
1859 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1860 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1861 return thread_pool_submit_co(pool
, func
, arg
);
1864 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1865 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1867 BDRVRawState
*s
= bs
->opaque
;
1868 RawPosixAIOData acb
;
1870 if (fd_open(bs
) < 0)
1874 * Check if the underlying device requires requests to be aligned,
1875 * and if the request we are trying to submit is aligned or not.
1876 * If this is the case tell the low-level driver that it needs
1877 * to copy the buffer.
1879 if (s
->needs_alignment
) {
1880 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1881 type
|= QEMU_AIO_MISALIGNED
;
1882 #ifdef CONFIG_LINUX_AIO
1883 } else if (s
->use_linux_aio
) {
1884 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1885 assert(qiov
->size
== bytes
);
1886 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1891 acb
= (RawPosixAIOData
) {
1893 .aio_fildes
= s
->fd
,
1895 .aio_offset
= offset
,
1896 .aio_nbytes
= bytes
,
1903 assert(qiov
->size
== bytes
);
1904 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1907 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1908 uint64_t bytes
, QEMUIOVector
*qiov
,
1911 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1914 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1915 uint64_t bytes
, QEMUIOVector
*qiov
,
1919 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1922 static void raw_aio_plug(BlockDriverState
*bs
)
1924 #ifdef CONFIG_LINUX_AIO
1925 BDRVRawState
*s
= bs
->opaque
;
1926 if (s
->use_linux_aio
) {
1927 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1928 laio_io_plug(bs
, aio
);
1933 static void raw_aio_unplug(BlockDriverState
*bs
)
1935 #ifdef CONFIG_LINUX_AIO
1936 BDRVRawState
*s
= bs
->opaque
;
1937 if (s
->use_linux_aio
) {
1938 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1939 laio_io_unplug(bs
, aio
);
1944 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1946 BDRVRawState
*s
= bs
->opaque
;
1947 RawPosixAIOData acb
;
1955 acb
= (RawPosixAIOData
) {
1957 .aio_fildes
= s
->fd
,
1958 .aio_type
= QEMU_AIO_FLUSH
,
1961 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
1964 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1965 AioContext
*new_context
)
1967 #ifdef CONFIG_LINUX_AIO
1968 BDRVRawState
*s
= bs
->opaque
;
1969 if (s
->use_linux_aio
) {
1971 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1972 error_reportf_err(local_err
, "Unable to use native AIO, "
1973 "falling back to thread pool: ");
1974 s
->use_linux_aio
= false;
1980 static void raw_close(BlockDriverState
*bs
)
1982 BDRVRawState
*s
= bs
->opaque
;
1991 * Truncates the given regular file @fd to @offset and, when growing, fills the
1992 * new space according to @prealloc.
1994 * Returns: 0 on success, -errno on failure.
1996 static int coroutine_fn
1997 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
1998 PreallocMode prealloc
, Error
**errp
)
2000 RawPosixAIOData acb
;
2002 acb
= (RawPosixAIOData
) {
2005 .aio_type
= QEMU_AIO_TRUNCATE
,
2006 .aio_offset
= offset
,
2008 .prealloc
= prealloc
,
2013 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2016 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2017 PreallocMode prealloc
, Error
**errp
)
2019 BDRVRawState
*s
= bs
->opaque
;
2023 if (fstat(s
->fd
, &st
)) {
2025 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2029 if (S_ISREG(st
.st_mode
)) {
2030 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2033 if (prealloc
!= PREALLOC_MODE_OFF
) {
2034 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2035 "non-regular file", PreallocMode_str(prealloc
));
2039 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2040 if (offset
> raw_getlength(bs
)) {
2041 error_setg(errp
, "Cannot grow device files");
2045 error_setg(errp
, "Resizing this file is not supported");
2053 static int64_t raw_getlength(BlockDriverState
*bs
)
2055 BDRVRawState
*s
= bs
->opaque
;
2061 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2062 struct disklabel dl
;
2064 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2066 return (uint64_t)dl
.d_secsize
*
2067 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2071 #elif defined(__NetBSD__)
2072 static int64_t raw_getlength(BlockDriverState
*bs
)
2074 BDRVRawState
*s
= bs
->opaque
;
2080 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2081 struct dkwedge_info dkw
;
2083 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2084 return dkw
.dkw_size
* 512;
2086 struct disklabel dl
;
2088 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2090 return (uint64_t)dl
.d_secsize
*
2091 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2096 #elif defined(__sun__)
2097 static int64_t raw_getlength(BlockDriverState
*bs
)
2099 BDRVRawState
*s
= bs
->opaque
;
2100 struct dk_minfo minfo
;
2110 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2112 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2114 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2118 * There are reports that lseek on some devices fails, but
2119 * irc discussion said that contingency on contingency was overkill.
2121 size
= lseek(s
->fd
, 0, SEEK_END
);
2127 #elif defined(CONFIG_BSD)
2128 static int64_t raw_getlength(BlockDriverState
*bs
)
2130 BDRVRawState
*s
= bs
->opaque
;
2134 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2143 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2146 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2147 #ifdef DIOCGMEDIASIZE
2148 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2149 #elif defined(DIOCGPART)
2152 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2153 size
= pi
.media_size
;
2159 #if defined(__APPLE__) && defined(__MACH__)
2161 uint64_t sectors
= 0;
2162 uint32_t sector_size
= 0;
2164 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2165 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2166 size
= sectors
* sector_size
;
2168 size
= lseek(fd
, 0LL, SEEK_END
);
2175 size
= lseek(fd
, 0LL, SEEK_END
);
2180 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2183 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2184 if (size
== 2048LL * (unsigned)-1)
2186 /* XXX no disc? maybe we need to reopen... */
2187 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2194 size
= lseek(fd
, 0, SEEK_END
);
2202 static int64_t raw_getlength(BlockDriverState
*bs
)
2204 BDRVRawState
*s
= bs
->opaque
;
2213 size
= lseek(s
->fd
, 0, SEEK_END
);
2221 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2224 BDRVRawState
*s
= bs
->opaque
;
2226 if (fstat(s
->fd
, &st
) < 0) {
2229 return (int64_t)st
.st_blocks
* 512;
2232 static int coroutine_fn
2233 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2235 BlockdevCreateOptionsFile
*file_opts
;
2236 Error
*local_err
= NULL
;
2238 uint64_t perm
, shared
;
2241 /* Validate options and set default values */
2242 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2243 file_opts
= &options
->u
.file
;
2245 if (!file_opts
->has_nocow
) {
2246 file_opts
->nocow
= false;
2248 if (!file_opts
->has_preallocation
) {
2249 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2253 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2256 error_setg_errno(errp
, -result
, "Could not create file");
2260 /* Take permissions: We want to discard everything, so we need
2261 * BLK_PERM_WRITE; and truncation to the desired size requires
2263 * On the other hand, we cannot share the RESIZE permission
2264 * because we promise that after this function, the file has the
2265 * size given in the options. If someone else were to resize it
2266 * concurrently, we could not guarantee that.
2267 * Note that after this function, we can no longer guarantee that
2268 * the file is not touched by a third party, so it may be resized
2270 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2271 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2273 /* Step one: Take locks */
2274 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2279 /* Step two: Check that nobody else has taken conflicting locks */
2280 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2282 error_append_hint(errp
,
2283 "Is another process using the image [%s]?\n",
2284 file_opts
->filename
);
2288 /* Clear the file by truncating it to 0 */
2289 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2294 if (file_opts
->nocow
) {
2296 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2297 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2298 * will be ignored since any failure of this operation should not
2299 * block the left work.
2302 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2303 attr
|= FS_NOCOW_FL
;
2304 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2309 /* Resize and potentially preallocate the file to the desired
2311 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2312 file_opts
->preallocation
, errp
);
2318 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2320 /* The above call should not fail, and if it does, that does
2321 * not mean the whole creation operation has failed. So
2322 * report it the user for their convenience, but do not report
2323 * it to the caller. */
2324 warn_report_err(local_err
);
2328 if (qemu_close(fd
) != 0 && result
== 0) {
2330 error_setg_errno(errp
, -result
, "Could not close the new file");
2336 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2339 BlockdevCreateOptions options
;
2340 int64_t total_size
= 0;
2342 PreallocMode prealloc
;
2344 Error
*local_err
= NULL
;
2346 /* Skip file: protocol prefix */
2347 strstart(filename
, "file:", &filename
);
2349 /* Read out options */
2350 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2352 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2353 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2354 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2355 PREALLOC_MODE_OFF
, &local_err
);
2358 error_propagate(errp
, local_err
);
2362 options
= (BlockdevCreateOptions
) {
2363 .driver
= BLOCKDEV_DRIVER_FILE
,
2365 .filename
= (char *) filename
,
2367 .has_preallocation
= true,
2368 .preallocation
= prealloc
,
2373 return raw_co_create(&options
, errp
);
2377 * Find allocation range in @bs around offset @start.
2378 * May change underlying file descriptor's file offset.
2379 * If @start is not in a hole, store @start in @data, and the
2380 * beginning of the next hole in @hole, and return 0.
2381 * If @start is in a non-trailing hole, store @start in @hole and the
2382 * beginning of the next non-hole in @data, and return 0.
2383 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2384 * If we can't find out, return a negative errno other than -ENXIO.
2386 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2387 off_t
*data
, off_t
*hole
)
2389 #if defined SEEK_HOLE && defined SEEK_DATA
2390 BDRVRawState
*s
= bs
->opaque
;
2395 * D1. offs == start: start is in data
2396 * D2. offs > start: start is in a hole, next data at offs
2397 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2398 * or start is beyond EOF
2399 * If the latter happens, the file has been truncated behind
2400 * our back since we opened it. All bets are off then.
2401 * Treating like a trailing hole is simplest.
2402 * D4. offs < 0, errno != ENXIO: we learned nothing
2404 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2406 return -errno
; /* D3 or D4 */
2410 /* This is not a valid return by lseek(). We are safe to just return
2411 * -EIO in this case, and we'll treat it like D4. */
2416 /* D2: in hole, next data at offs */
2422 /* D1: in data, end not yet known */
2426 * H1. offs == start: start is in a hole
2427 * If this happens here, a hole has been dug behind our back
2428 * since the previous lseek().
2429 * H2. offs > start: either start is in data, next hole at offs,
2430 * or start is in trailing hole, EOF at offs
2431 * Linux treats trailing holes like any other hole: offs ==
2432 * start. Solaris seeks to EOF instead: offs > start (blech).
2433 * If that happens here, a hole has been dug behind our back
2434 * since the previous lseek().
2435 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2436 * If this happens, the file has been truncated behind our
2437 * back since we opened it. Treat it like a trailing hole.
2438 * H4. offs < 0, errno != ENXIO: we learned nothing
2439 * Pretend we know nothing at all, i.e. "forget" about D1.
2441 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2443 return -errno
; /* D1 and (H3 or H4) */
2447 /* This is not a valid return by lseek(). We are safe to just return
2448 * -EIO in this case, and we'll treat it like H4. */
2454 * D1 and H2: either in data, next hole at offs, or it was in
2455 * data but is now in a trailing hole. In the latter case,
2456 * all bets are off. Treating it as if it there was data all
2457 * the way to EOF is safe, so simply do that.
2472 * Returns the allocation status of the specified offset.
2474 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2476 * 'pnum' is set to the number of bytes (including and immediately following
2477 * the specified offset) that are known to be in the same
2478 * allocated/unallocated state.
2480 * 'bytes' is the max value 'pnum' should be set to.
2482 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2485 int64_t bytes
, int64_t *pnum
,
2487 BlockDriverState
**file
)
2489 off_t data
= 0, hole
= 0;
2492 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2503 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2506 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2507 if (ret
== -ENXIO
) {
2510 ret
= BDRV_BLOCK_ZERO
;
2511 } else if (ret
< 0) {
2512 /* No info available, so pretend there are no holes */
2514 ret
= BDRV_BLOCK_DATA
;
2515 } else if (data
== offset
) {
2516 /* On a data extent, compute bytes to the end of the extent,
2517 * possibly including a partial sector at EOF. */
2518 *pnum
= MIN(bytes
, hole
- offset
);
2521 * We are not allowed to return partial sectors, though, so
2522 * round up if necessary.
2524 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2525 int64_t file_length
= raw_getlength(bs
);
2526 if (file_length
> 0) {
2527 /* Ignore errors, this is just a safeguard */
2528 assert(hole
== file_length
);
2530 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2533 ret
= BDRV_BLOCK_DATA
;
2535 /* On a hole, compute bytes to the beginning of the next extent. */
2536 assert(hole
== offset
);
2537 *pnum
= MIN(bytes
, data
- offset
);
2538 ret
= BDRV_BLOCK_ZERO
;
2542 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2545 #if defined(__linux__)
2546 /* Verify that the file is not in the page cache */
2547 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2549 const size_t window_size
= 128 * 1024 * 1024;
2550 BDRVRawState
*s
= bs
->opaque
;
2551 void *window
= NULL
;
2558 /* mincore(2) page status information requires 1 byte per page */
2559 page_size
= sysconf(_SC_PAGESIZE
);
2560 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2562 end
= raw_getlength(bs
);
2564 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2571 /* Unmap previous window if size has changed */
2572 new_length
= MIN(end
- offset
, window_size
);
2573 if (new_length
!= length
) {
2574 munmap(window
, length
);
2579 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2581 if (new_window
== MAP_FAILED
) {
2582 error_setg_errno(errp
, errno
, "mmap failed");
2586 window
= new_window
;
2587 length
= new_length
;
2589 ret
= mincore(window
, length
, vec
);
2591 error_setg_errno(errp
, errno
, "mincore failed");
2595 vec_end
= DIV_ROUND_UP(length
, page_size
);
2596 for (i
= 0; i
< vec_end
; i
++) {
2598 error_setg(errp
, "page cache still in use!");
2605 munmap(window
, length
);
2610 #endif /* __linux__ */
2612 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2615 BDRVRawState
*s
= bs
->opaque
;
2620 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2624 if (!s
->drop_cache
) {
2628 if (s
->open_flags
& O_DIRECT
) {
2629 return; /* No host kernel page cache */
2632 #if defined(__linux__)
2633 /* This sets the scene for the next syscall... */
2634 ret
= bdrv_co_flush(bs
);
2636 error_setg_errno(errp
, -ret
, "flush failed");
2640 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2641 * process. These limitations are okay because we just fsynced the file,
2642 * we don't use mmap, and the file should not be in use by other processes.
2644 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2645 if (ret
!= 0) { /* the return value is a positive errno */
2646 error_setg_errno(errp
, ret
, "fadvise failed");
2650 if (s
->check_cache_dropped
) {
2651 check_cache_dropped(bs
, errp
);
2653 #else /* __linux__ */
2654 /* Do nothing. Live migration to a remote host with cache.direct=off is
2655 * unsupported on other host operating systems. Cache consistency issues
2656 * may occur but no error is reported here, partly because that's the
2657 * historical behavior and partly because it's hard to differentiate valid
2658 * configurations that should not cause errors.
2660 #endif /* !__linux__ */
2663 static coroutine_fn
int
2664 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2666 BDRVRawState
*s
= bs
->opaque
;
2667 RawPosixAIOData acb
;
2669 acb
= (RawPosixAIOData
) {
2671 .aio_fildes
= s
->fd
,
2672 .aio_type
= QEMU_AIO_DISCARD
,
2673 .aio_offset
= offset
,
2674 .aio_nbytes
= bytes
,
2678 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2681 return raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2684 static coroutine_fn
int
2685 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2687 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2690 static int coroutine_fn
2691 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2692 BdrvRequestFlags flags
, bool blkdev
)
2694 BDRVRawState
*s
= bs
->opaque
;
2695 RawPosixAIOData acb
;
2696 ThreadPoolFunc
*handler
;
2698 acb
= (RawPosixAIOData
) {
2700 .aio_fildes
= s
->fd
,
2701 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2702 .aio_offset
= offset
,
2703 .aio_nbytes
= bytes
,
2707 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2709 if (flags
& BDRV_REQ_NO_FALLBACK
) {
2710 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
2713 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2714 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2715 handler
= handle_aiocb_write_zeroes_unmap
;
2717 handler
= handle_aiocb_write_zeroes
;
2720 return raw_thread_pool_submit(bs
, handler
, &acb
);
2723 static int coroutine_fn
raw_co_pwrite_zeroes(
2724 BlockDriverState
*bs
, int64_t offset
,
2725 int bytes
, BdrvRequestFlags flags
)
2727 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2730 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2732 BDRVRawState
*s
= bs
->opaque
;
2734 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2738 static QemuOptsList raw_create_opts
= {
2739 .name
= "raw-create-opts",
2740 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2743 .name
= BLOCK_OPT_SIZE
,
2744 .type
= QEMU_OPT_SIZE
,
2745 .help
= "Virtual disk size"
2748 .name
= BLOCK_OPT_NOCOW
,
2749 .type
= QEMU_OPT_BOOL
,
2750 .help
= "Turn off copy-on-write (valid only on btrfs)"
2753 .name
= BLOCK_OPT_PREALLOC
,
2754 .type
= QEMU_OPT_STRING
,
2755 .help
= "Preallocation mode (allowed values: off"
2756 #ifdef CONFIG_POSIX_FALLOCATE
2761 { /* end of list */ }
2765 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2768 BDRVRawState
*s
= bs
->opaque
;
2769 BDRVRawReopenState
*rs
= NULL
;
2773 if (s
->perm_change_fd
) {
2775 * In the context of reopen, this function may be called several times
2776 * (directly and recursively while change permissions of the parent).
2777 * This is even true for children that don't inherit from the original
2778 * reopen node, so s->reopen_state is not set.
2780 * Ignore all but the first call.
2785 if (s
->reopen_state
) {
2786 /* We already have a new file descriptor to set permissions for */
2787 assert(s
->reopen_state
->perm
== perm
);
2788 assert(s
->reopen_state
->shared_perm
== shared
);
2789 rs
= s
->reopen_state
->opaque
;
2790 s
->perm_change_fd
= rs
->fd
;
2791 s
->perm_change_flags
= rs
->open_flags
;
2793 /* We may need a new fd if auto-read-only switches the mode */
2794 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
2798 } else if (ret
!= s
->fd
) {
2799 s
->perm_change_fd
= ret
;
2800 s
->perm_change_flags
= open_flags
;
2804 /* Prepare permissions on old fd to avoid conflicts between old and new,
2805 * but keep everything locked that new will need. */
2806 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2811 /* Copy locks to the new fd */
2812 if (s
->perm_change_fd
) {
2813 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
2816 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2823 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2824 qemu_close(s
->perm_change_fd
);
2826 s
->perm_change_fd
= 0;
2830 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2832 BDRVRawState
*s
= bs
->opaque
;
2834 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
2835 * called after .bdrv_reopen_commit) */
2836 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
2838 s
->fd
= s
->perm_change_fd
;
2839 s
->open_flags
= s
->perm_change_flags
;
2841 s
->perm_change_fd
= 0;
2843 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2845 s
->shared_perm
= shared
;
2848 static void raw_abort_perm_update(BlockDriverState
*bs
)
2850 BDRVRawState
*s
= bs
->opaque
;
2852 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
2853 * the file descriptor. */
2854 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2855 qemu_close(s
->perm_change_fd
);
2857 s
->perm_change_fd
= 0;
2859 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2862 static int coroutine_fn
raw_co_copy_range_from(
2863 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2864 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2865 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2867 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2868 read_flags
, write_flags
);
2871 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2873 uint64_t src_offset
,
2875 uint64_t dst_offset
,
2877 BdrvRequestFlags read_flags
,
2878 BdrvRequestFlags write_flags
)
2880 RawPosixAIOData acb
;
2881 BDRVRawState
*s
= bs
->opaque
;
2882 BDRVRawState
*src_s
;
2884 assert(dst
->bs
== bs
);
2885 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2889 src_s
= src
->bs
->opaque
;
2890 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2894 acb
= (RawPosixAIOData
) {
2896 .aio_type
= QEMU_AIO_COPY_RANGE
,
2897 .aio_fildes
= src_s
->fd
,
2898 .aio_offset
= src_offset
,
2899 .aio_nbytes
= bytes
,
2902 .aio_offset2
= dst_offset
,
2906 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
2909 BlockDriver bdrv_file
= {
2910 .format_name
= "file",
2911 .protocol_name
= "file",
2912 .instance_size
= sizeof(BDRVRawState
),
2913 .bdrv_needs_filename
= true,
2914 .bdrv_probe
= NULL
, /* no probe for protocols */
2915 .bdrv_parse_filename
= raw_parse_filename
,
2916 .bdrv_file_open
= raw_open
,
2917 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2918 .bdrv_reopen_commit
= raw_reopen_commit
,
2919 .bdrv_reopen_abort
= raw_reopen_abort
,
2920 .bdrv_close
= raw_close
,
2921 .bdrv_co_create
= raw_co_create
,
2922 .bdrv_co_create_opts
= raw_co_create_opts
,
2923 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2924 .bdrv_has_zero_init_truncate
= bdrv_has_zero_init_1
,
2925 .bdrv_co_block_status
= raw_co_block_status
,
2926 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2927 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2929 .bdrv_co_preadv
= raw_co_preadv
,
2930 .bdrv_co_pwritev
= raw_co_pwritev
,
2931 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2932 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2933 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2934 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2935 .bdrv_refresh_limits
= raw_refresh_limits
,
2936 .bdrv_io_plug
= raw_aio_plug
,
2937 .bdrv_io_unplug
= raw_aio_unplug
,
2938 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2940 .bdrv_co_truncate
= raw_co_truncate
,
2941 .bdrv_getlength
= raw_getlength
,
2942 .bdrv_get_info
= raw_get_info
,
2943 .bdrv_get_allocated_file_size
2944 = raw_get_allocated_file_size
,
2945 .bdrv_check_perm
= raw_check_perm
,
2946 .bdrv_set_perm
= raw_set_perm
,
2947 .bdrv_abort_perm_update
= raw_abort_perm_update
,
2948 .create_opts
= &raw_create_opts
,
2949 .mutable_opts
= mutable_opts
,
2952 /***********************************************/
2955 #if defined(__APPLE__) && defined(__MACH__)
2956 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
2957 CFIndex maxPathSize
, int flags
);
2958 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
2960 kern_return_t kernResult
= KERN_FAILURE
;
2961 mach_port_t masterPort
;
2962 CFMutableDictionaryRef classesToMatch
;
2963 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
2964 char *mediaType
= NULL
;
2966 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
2967 if ( KERN_SUCCESS
!= kernResult
) {
2968 printf( "IOMasterPort returned %d\n", kernResult
);
2972 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
2973 classesToMatch
= IOServiceMatching(matching_array
[index
]);
2974 if (classesToMatch
== NULL
) {
2975 error_report("IOServiceMatching returned NULL for %s",
2976 matching_array
[index
]);
2979 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
2981 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
2983 if (kernResult
!= KERN_SUCCESS
) {
2984 error_report("Note: IOServiceGetMatchingServices returned %d",
2989 /* If a match was found, leave the loop */
2990 if (*mediaIterator
!= 0) {
2991 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
2992 mediaType
= g_strdup(matching_array
[index
]);
2999 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3000 CFIndex maxPathSize
, int flags
)
3002 io_object_t nextMedia
;
3003 kern_return_t kernResult
= KERN_FAILURE
;
3005 nextMedia
= IOIteratorNext( mediaIterator
);
3008 CFTypeRef bsdPathAsCFString
;
3009 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3010 if ( bsdPathAsCFString
) {
3011 size_t devPathLength
;
3012 strcpy( bsdPath
, _PATH_DEV
);
3013 if (flags
& BDRV_O_NOCACHE
) {
3014 strcat(bsdPath
, "r");
3016 devPathLength
= strlen( bsdPath
);
3017 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3018 kernResult
= KERN_SUCCESS
;
3020 CFRelease( bsdPathAsCFString
);
3022 IOObjectRelease( nextMedia
);
3028 /* Sets up a real cdrom for use in QEMU */
3029 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3031 int index
, num_of_test_partitions
= 2, fd
;
3032 char test_partition
[MAXPATHLEN
];
3033 bool partition_found
= false;
3035 /* look for a working partition */
3036 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3037 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3039 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
3041 partition_found
= true;
3047 /* if a working partition on the device was not found */
3048 if (partition_found
== false) {
3049 error_setg(errp
, "Failed to find a working partition on disc");
3051 trace_file_setup_cdrom(test_partition
);
3052 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3054 return partition_found
;
3057 /* Prints directions on mounting and unmounting a device */
3058 static void print_unmounting_directions(const char *file_name
)
3060 error_report("If device %s is mounted on the desktop, unmount"
3061 " it first before using it in QEMU", file_name
);
3062 error_report("Command to unmount device: diskutil unmountDisk %s",
3064 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3067 #endif /* defined(__APPLE__) && defined(__MACH__) */
3069 static int hdev_probe_device(const char *filename
)
3073 /* allow a dedicated CD-ROM driver to match with a higher priority */
3074 if (strstart(filename
, "/dev/cdrom", NULL
))
3077 if (stat(filename
, &st
) >= 0 &&
3078 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3085 static int check_hdev_writable(BDRVRawState
*s
)
3087 #if defined(BLKROGET)
3088 /* Linux block devices can be configured "read-only" using blockdev(8).
3089 * This is independent of device node permissions and therefore open(2)
3090 * with O_RDWR succeeds. Actual writes fail with EPERM.
3092 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3093 * check for read-only block devices so that Linux block devices behave
3099 if (fstat(s
->fd
, &st
)) {
3103 if (!S_ISBLK(st
.st_mode
)) {
3107 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
3114 #endif /* defined(BLKROGET) */
3118 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3121 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3124 static bool hdev_is_sg(BlockDriverState
*bs
)
3127 #if defined(__linux__)
3129 BDRVRawState
*s
= bs
->opaque
;
3131 struct sg_scsi_id scsiid
;
3135 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3139 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3144 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3146 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3155 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3158 BDRVRawState
*s
= bs
->opaque
;
3159 Error
*local_err
= NULL
;
3162 #if defined(__APPLE__) && defined(__MACH__)
3164 * Caution: while qdict_get_str() is fine, getting non-string types
3165 * would require more care. When @options come from -blockdev or
3166 * blockdev_add, its members are typed according to the QAPI
3167 * schema, but when they come from -drive, they're all QString.
3169 const char *filename
= qdict_get_str(options
, "filename");
3170 char bsd_path
[MAXPATHLEN
] = "";
3171 bool error_occurred
= false;
3173 /* If using a real cdrom */
3174 if (strcmp(filename
, "/dev/cdrom") == 0) {
3175 char *mediaType
= NULL
;
3176 kern_return_t ret_val
;
3177 io_iterator_t mediaIterator
= 0;
3179 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3180 if (mediaType
== NULL
) {
3181 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3183 error_occurred
= true;
3184 goto hdev_open_Mac_error
;
3187 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3188 if (ret_val
!= KERN_SUCCESS
) {
3189 error_setg(errp
, "Could not get BSD path for optical drive");
3190 error_occurred
= true;
3191 goto hdev_open_Mac_error
;
3194 /* If a real optical drive was not found */
3195 if (bsd_path
[0] == '\0') {
3196 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3197 error_occurred
= true;
3198 goto hdev_open_Mac_error
;
3201 /* If using a cdrom disc and finding a partition on the disc failed */
3202 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3203 setup_cdrom(bsd_path
, errp
) == false) {
3204 print_unmounting_directions(bsd_path
);
3205 error_occurred
= true;
3206 goto hdev_open_Mac_error
;
3209 qdict_put_str(options
, "filename", bsd_path
);
3211 hdev_open_Mac_error
:
3213 if (mediaIterator
) {
3214 IOObjectRelease(mediaIterator
);
3216 if (error_occurred
) {
3220 #endif /* defined(__APPLE__) && defined(__MACH__) */
3222 s
->type
= FTYPE_FILE
;
3224 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3226 error_propagate(errp
, local_err
);
3227 #if defined(__APPLE__) && defined(__MACH__)
3229 filename
= bsd_path
;
3231 /* if a physical device experienced an error while being opened */
3232 if (strncmp(filename
, "/dev/", 5) == 0) {
3233 print_unmounting_directions(filename
);
3235 #endif /* defined(__APPLE__) && defined(__MACH__) */
3239 /* Since this does ioctl the device must be already opened */
3240 bs
->sg
= hdev_is_sg(bs
);
3242 if (flags
& BDRV_O_RDWR
) {
3243 ret
= check_hdev_writable(s
);
3246 error_setg_errno(errp
, -ret
, "The device is not writable");
3254 #if defined(__linux__)
3255 static int coroutine_fn
3256 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3258 BDRVRawState
*s
= bs
->opaque
;
3259 RawPosixAIOData acb
;
3267 if (req
== SG_IO
&& s
->pr_mgr
) {
3268 struct sg_io_hdr
*io_hdr
= buf
;
3269 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3270 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3271 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3276 acb
= (RawPosixAIOData
) {
3278 .aio_type
= QEMU_AIO_IOCTL
,
3279 .aio_fildes
= s
->fd
,
3287 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3291 static int fd_open(BlockDriverState
*bs
)
3293 BDRVRawState
*s
= bs
->opaque
;
3295 /* this is just to ensure s->fd is sane (its called by io ops) */
3301 static coroutine_fn
int
3302 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3310 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3313 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3314 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3323 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3326 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3331 struct stat stat_buf
;
3332 int64_t total_size
= 0;
3335 /* This function is used by both protocol block drivers and therefore either
3336 * of these prefixes may be given.
3337 * The return value has to be stored somewhere, otherwise this is an error
3338 * due to -Werror=unused-value. */
3340 strstart(filename
, "host_device:", &filename
) ||
3341 strstart(filename
, "host_cdrom:" , &filename
);
3345 ret
= raw_normalize_devicepath(&filename
, errp
);
3350 /* Read out options */
3351 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3354 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3357 error_setg_errno(errp
, -ret
, "Could not open device");
3361 if (fstat(fd
, &stat_buf
) < 0) {
3363 error_setg_errno(errp
, -ret
, "Could not stat device");
3364 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3366 "The given file is neither a block nor a character device");
3368 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3369 error_setg(errp
, "Device is too small");
3373 if (!ret
&& total_size
) {
3374 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3375 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3376 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3379 ret
= qemu_write_full(fd
, buf
, zero_size
);
3380 ret
= ret
== zero_size
? 0 : -errno
;
3387 static BlockDriver bdrv_host_device
= {
3388 .format_name
= "host_device",
3389 .protocol_name
= "host_device",
3390 .instance_size
= sizeof(BDRVRawState
),
3391 .bdrv_needs_filename
= true,
3392 .bdrv_probe_device
= hdev_probe_device
,
3393 .bdrv_parse_filename
= hdev_parse_filename
,
3394 .bdrv_file_open
= hdev_open
,
3395 .bdrv_close
= raw_close
,
3396 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3397 .bdrv_reopen_commit
= raw_reopen_commit
,
3398 .bdrv_reopen_abort
= raw_reopen_abort
,
3399 .bdrv_co_create_opts
= hdev_co_create_opts
,
3400 .create_opts
= &raw_create_opts
,
3401 .mutable_opts
= mutable_opts
,
3402 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3403 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3405 .bdrv_co_preadv
= raw_co_preadv
,
3406 .bdrv_co_pwritev
= raw_co_pwritev
,
3407 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3408 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3409 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3410 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3411 .bdrv_refresh_limits
= raw_refresh_limits
,
3412 .bdrv_io_plug
= raw_aio_plug
,
3413 .bdrv_io_unplug
= raw_aio_unplug
,
3414 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3416 .bdrv_co_truncate
= raw_co_truncate
,
3417 .bdrv_getlength
= raw_getlength
,
3418 .bdrv_get_info
= raw_get_info
,
3419 .bdrv_get_allocated_file_size
3420 = raw_get_allocated_file_size
,
3421 .bdrv_check_perm
= raw_check_perm
,
3422 .bdrv_set_perm
= raw_set_perm
,
3423 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3424 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3425 .bdrv_probe_geometry
= hdev_probe_geometry
,
3427 /* generic scsi device */
3429 .bdrv_co_ioctl
= hdev_co_ioctl
,
3433 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3434 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3437 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3442 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3445 BDRVRawState
*s
= bs
->opaque
;
3449 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3450 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3453 static int cdrom_probe_device(const char *filename
)
3459 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3463 ret
= fstat(fd
, &st
);
3464 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3468 /* Attempt to detect via a CDROM specific ioctl */
3469 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3479 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3481 BDRVRawState
*s
= bs
->opaque
;
3484 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3485 return ret
== CDS_DISC_OK
;
3488 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3490 BDRVRawState
*s
= bs
->opaque
;
3493 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3494 perror("CDROMEJECT");
3496 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3497 perror("CDROMEJECT");
3501 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3503 BDRVRawState
*s
= bs
->opaque
;
3505 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3507 * Note: an error can happen if the distribution automatically
3510 /* perror("CDROM_LOCKDOOR"); */
3514 static BlockDriver bdrv_host_cdrom
= {
3515 .format_name
= "host_cdrom",
3516 .protocol_name
= "host_cdrom",
3517 .instance_size
= sizeof(BDRVRawState
),
3518 .bdrv_needs_filename
= true,
3519 .bdrv_probe_device
= cdrom_probe_device
,
3520 .bdrv_parse_filename
= cdrom_parse_filename
,
3521 .bdrv_file_open
= cdrom_open
,
3522 .bdrv_close
= raw_close
,
3523 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3524 .bdrv_reopen_commit
= raw_reopen_commit
,
3525 .bdrv_reopen_abort
= raw_reopen_abort
,
3526 .bdrv_co_create_opts
= hdev_co_create_opts
,
3527 .create_opts
= &raw_create_opts
,
3528 .mutable_opts
= mutable_opts
,
3529 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3532 .bdrv_co_preadv
= raw_co_preadv
,
3533 .bdrv_co_pwritev
= raw_co_pwritev
,
3534 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
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 .has_variable_length
= true,
3543 .bdrv_get_allocated_file_size
3544 = raw_get_allocated_file_size
,
3546 /* removable device support */
3547 .bdrv_is_inserted
= cdrom_is_inserted
,
3548 .bdrv_eject
= cdrom_eject
,
3549 .bdrv_lock_medium
= cdrom_lock_medium
,
3551 /* generic scsi device */
3552 .bdrv_co_ioctl
= hdev_co_ioctl
,
3554 #endif /* __linux__ */
3556 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3557 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3560 BDRVRawState
*s
= bs
->opaque
;
3561 Error
*local_err
= NULL
;
3566 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3568 error_propagate(errp
, local_err
);
3572 /* make sure the door isn't locked at this time */
3573 ioctl(s
->fd
, CDIOCALLOW
);
3577 static int cdrom_probe_device(const char *filename
)
3579 if (strstart(filename
, "/dev/cd", NULL
) ||
3580 strstart(filename
, "/dev/acd", NULL
))
3585 static int cdrom_reopen(BlockDriverState
*bs
)
3587 BDRVRawState
*s
= bs
->opaque
;
3591 * Force reread of possibly changed/newly loaded disc,
3592 * FreeBSD seems to not notice sometimes...
3596 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3603 /* make sure the door isn't locked at this time */
3604 ioctl(s
->fd
, CDIOCALLOW
);
3608 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3610 return raw_getlength(bs
) > 0;
3613 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3615 BDRVRawState
*s
= bs
->opaque
;
3620 (void) ioctl(s
->fd
, CDIOCALLOW
);
3623 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3624 perror("CDIOCEJECT");
3626 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3627 perror("CDIOCCLOSE");
3633 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3635 BDRVRawState
*s
= bs
->opaque
;
3639 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3641 * Note: an error can happen if the distribution automatically
3644 /* perror("CDROM_LOCKDOOR"); */
3648 static BlockDriver bdrv_host_cdrom
= {
3649 .format_name
= "host_cdrom",
3650 .protocol_name
= "host_cdrom",
3651 .instance_size
= sizeof(BDRVRawState
),
3652 .bdrv_needs_filename
= true,
3653 .bdrv_probe_device
= cdrom_probe_device
,
3654 .bdrv_parse_filename
= cdrom_parse_filename
,
3655 .bdrv_file_open
= cdrom_open
,
3656 .bdrv_close
= raw_close
,
3657 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3658 .bdrv_reopen_commit
= raw_reopen_commit
,
3659 .bdrv_reopen_abort
= raw_reopen_abort
,
3660 .bdrv_co_create_opts
= hdev_co_create_opts
,
3661 .create_opts
= &raw_create_opts
,
3662 .mutable_opts
= mutable_opts
,
3664 .bdrv_co_preadv
= raw_co_preadv
,
3665 .bdrv_co_pwritev
= raw_co_pwritev
,
3666 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3667 .bdrv_refresh_limits
= raw_refresh_limits
,
3668 .bdrv_io_plug
= raw_aio_plug
,
3669 .bdrv_io_unplug
= raw_aio_unplug
,
3670 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3672 .bdrv_co_truncate
= raw_co_truncate
,
3673 .bdrv_getlength
= raw_getlength
,
3674 .has_variable_length
= true,
3675 .bdrv_get_allocated_file_size
3676 = raw_get_allocated_file_size
,
3678 /* removable device support */
3679 .bdrv_is_inserted
= cdrom_is_inserted
,
3680 .bdrv_eject
= cdrom_eject
,
3681 .bdrv_lock_medium
= cdrom_lock_medium
,
3683 #endif /* __FreeBSD__ */
3685 static void bdrv_file_init(void)
3688 * Register all the drivers. Note that order is important, the driver
3689 * registered last will get probed first.
3691 bdrv_register(&bdrv_file
);
3692 bdrv_register(&bdrv_host_device
);
3694 bdrv_register(&bdrv_host_cdrom
);
3696 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3697 bdrv_register(&bdrv_host_cdrom
);
3701 block_init(bdrv_file_init
);