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
;
165 uint64_t discard_nb_ok
;
166 uint64_t discard_nb_failed
;
167 uint64_t discard_bytes_ok
;
173 typedef struct BDRVRawReopenState
{
177 bool check_cache_dropped
;
178 } BDRVRawReopenState
;
180 static int fd_open(BlockDriverState
*bs
);
181 static int64_t raw_getlength(BlockDriverState
*bs
);
183 typedef struct RawPosixAIOData
{
184 BlockDriverState
*bs
;
205 PreallocMode prealloc
;
211 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
212 static int cdrom_reopen(BlockDriverState
*bs
);
215 #if defined(__NetBSD__)
216 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
218 static char namebuf
[PATH_MAX
];
219 const char *dp
, *fname
;
223 dp
= strrchr(fname
, '/');
224 if (lstat(fname
, &sb
) < 0) {
225 error_setg_file_open(errp
, errno
, fname
);
229 if (!S_ISBLK(sb
.st_mode
)) {
234 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
236 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
237 (int)(dp
- fname
), fname
, dp
+ 1);
240 warn_report("%s is a block device, using %s", fname
, *filename
);
245 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
252 * Get logical block size via ioctl. On success store it in @sector_size_p.
254 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
256 unsigned int sector_size
;
257 bool success
= false;
261 static const unsigned long ioctl_list
[] = {
265 #ifdef DKIOCGETBLOCKSIZE
268 #ifdef DIOCGSECTORSIZE
273 /* Try a few ioctls to get the right size */
274 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
275 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
276 *sector_size_p
= sector_size
;
281 return success
? 0 : -errno
;
285 * Get physical block size of @fd.
286 * On success, store it in @blk_size and return 0.
287 * On failure, return -errno.
289 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
292 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
301 /* Check if read is allowed with given memory buffer and length.
303 * This function is used to check O_DIRECT memory buffer and request alignment.
305 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
307 ssize_t ret
= pread(fd
, buf
, len
, 0);
314 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
315 * other errors (e.g. real I/O error), which could happen on a failed
316 * drive, since we only care about probing alignment.
318 if (errno
!= EINVAL
) {
326 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
328 BDRVRawState
*s
= bs
->opaque
;
330 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
331 size_t alignments
[] = {1, 512, 1024, 2048, 4096};
333 /* For SCSI generic devices the alignment is not really used.
334 With buffered I/O, we don't have any restrictions. */
335 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
336 bs
->bl
.request_alignment
= 1;
341 bs
->bl
.request_alignment
= 0;
343 /* Let's try to use the logical blocksize for the alignment. */
344 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
345 bs
->bl
.request_alignment
= 0;
350 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
351 bs
->bl
.request_alignment
= da
.d_miniosz
;
352 /* The kernel returns wrong information for d_mem */
353 /* s->buf_align = da.d_mem; */
359 * If we could not get the sizes so far, we can only guess them. First try
360 * to detect request alignment, since it is more likely to succeed. Then
361 * try to detect buf_align, which cannot be detected in some cases (e.g.
362 * Gluster). If buf_align cannot be detected, we fallback to the value of
366 if (!bs
->bl
.request_alignment
) {
369 buf
= qemu_memalign(max_align
, max_align
);
370 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
371 align
= alignments
[i
];
372 if (raw_is_io_aligned(fd
, buf
, align
)) {
373 /* Fallback to safe value. */
374 bs
->bl
.request_alignment
= (align
!= 1) ? align
: max_align
;
384 buf
= qemu_memalign(max_align
, 2 * max_align
);
385 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
386 align
= alignments
[i
];
387 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
388 /* Fallback to request_alignment. */
389 s
->buf_align
= (align
!= 1) ? align
: bs
->bl
.request_alignment
;
396 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
397 error_setg(errp
, "Could not find working O_DIRECT alignment");
398 error_append_hint(errp
, "Try cache.direct=off\n");
402 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
404 bool read_write
= false;
405 assert(open_flags
!= NULL
);
407 *open_flags
|= O_BINARY
;
408 *open_flags
&= ~O_ACCMODE
;
410 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
411 read_write
= has_writers
;
412 } else if (bdrv_flags
& BDRV_O_RDWR
) {
417 *open_flags
|= O_RDWR
;
419 *open_flags
|= O_RDONLY
;
422 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
423 * and O_DIRECT for no caching. */
424 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
425 *open_flags
|= O_DIRECT
;
429 static void raw_parse_filename(const char *filename
, QDict
*options
,
432 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
435 static QemuOptsList raw_runtime_opts
= {
437 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
441 .type
= QEMU_OPT_STRING
,
442 .help
= "File name of the image",
446 .type
= QEMU_OPT_STRING
,
447 .help
= "host AIO implementation (threads, native)",
451 .type
= QEMU_OPT_STRING
,
452 .help
= "file locking mode (on/off/auto, default: auto)",
455 .name
= "pr-manager",
456 .type
= QEMU_OPT_STRING
,
457 .help
= "id of persistent reservation manager object (default: none)",
459 #if defined(__linux__)
461 .name
= "drop-cache",
462 .type
= QEMU_OPT_BOOL
,
463 .help
= "invalidate page cache during live migration (default: on)",
467 .name
= "x-check-cache-dropped",
468 .type
= QEMU_OPT_BOOL
,
469 .help
= "check that page cache was dropped on live migration (default: off)"
471 { /* end of list */ }
475 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
477 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
478 int bdrv_flags
, int open_flags
,
479 bool device
, Error
**errp
)
481 BDRVRawState
*s
= bs
->opaque
;
483 Error
*local_err
= NULL
;
484 const char *filename
= NULL
;
486 BlockdevAioOptions aio
, aio_default
;
491 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
492 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
494 error_propagate(errp
, local_err
);
499 filename
= qemu_opt_get(opts
, "filename");
501 ret
= raw_normalize_devicepath(&filename
, errp
);
506 aio_default
= (bdrv_flags
& BDRV_O_NATIVE_AIO
)
507 ? BLOCKDEV_AIO_OPTIONS_NATIVE
508 : BLOCKDEV_AIO_OPTIONS_THREADS
;
509 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
510 qemu_opt_get(opts
, "aio"),
511 aio_default
, &local_err
);
513 error_propagate(errp
, local_err
);
517 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
519 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
520 qemu_opt_get(opts
, "locking"),
521 ON_OFF_AUTO_AUTO
, &local_err
);
523 error_propagate(errp
, local_err
);
530 if (!qemu_has_ofd_lock()) {
531 warn_report("File lock requested but OFD locking syscall is "
532 "unavailable, falling back to POSIX file locks");
533 error_printf("Due to the implementation, locks can be lost "
537 case ON_OFF_AUTO_OFF
:
540 case ON_OFF_AUTO_AUTO
:
541 s
->use_lock
= qemu_has_ofd_lock();
547 str
= qemu_opt_get(opts
, "pr-manager");
549 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
551 error_propagate(errp
, local_err
);
557 s
->drop_cache
= qemu_opt_get_bool(opts
, "drop-cache", true);
558 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
561 s
->open_flags
= open_flags
;
562 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
565 fd
= qemu_open(filename
, s
->open_flags
, 0644);
566 ret
= fd
< 0 ? -errno
: 0;
569 error_setg_file_open(errp
, -ret
, filename
);
578 s
->shared_perm
= BLK_PERM_ALL
;
580 #ifdef CONFIG_LINUX_AIO
581 /* Currently Linux does AIO only for files opened with O_DIRECT */
582 if (s
->use_linux_aio
) {
583 if (!(s
->open_flags
& O_DIRECT
)) {
584 error_setg(errp
, "aio=native was specified, but it requires "
585 "cache.direct=on, which was not specified.");
589 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
590 error_prepend(errp
, "Unable to use native AIO: ");
595 if (s
->use_linux_aio
) {
596 error_setg(errp
, "aio=native was specified, but is not supported "
601 #endif /* !defined(CONFIG_LINUX_AIO) */
603 s
->has_discard
= true;
604 s
->has_write_zeroes
= true;
605 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
606 s
->needs_alignment
= true;
609 if (fstat(s
->fd
, &st
) < 0) {
611 error_setg_errno(errp
, errno
, "Could not stat file");
616 if (S_ISBLK(st
.st_mode
)) {
617 warn_report("Opening a block device as a file using the '%s' "
618 "driver is deprecated", bs
->drv
->format_name
);
619 } else if (S_ISCHR(st
.st_mode
)) {
620 warn_report("Opening a character device as a file using the '%s' "
621 "driver is deprecated", bs
->drv
->format_name
);
622 } else if (!S_ISREG(st
.st_mode
)) {
623 error_setg(errp
, "A regular file was expected by the '%s' driver, "
624 "but something else was given", bs
->drv
->format_name
);
628 s
->discard_zeroes
= true;
629 s
->has_fallocate
= true;
632 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
633 error_setg(errp
, "'%s' driver expects either "
634 "a character or block device", bs
->drv
->format_name
);
640 if (S_ISBLK(st
.st_mode
)) {
641 #ifdef BLKDISCARDZEROES
643 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
644 s
->discard_zeroes
= true;
648 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
649 * not rely on the contents of discarded blocks unless using O_DIRECT.
650 * Same for BLKZEROOUT.
652 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
653 s
->discard_zeroes
= false;
654 s
->has_write_zeroes
= false;
659 if (S_ISCHR(st
.st_mode
)) {
661 * The file is a char device (disk), which on FreeBSD isn't behind
662 * a pager, so force all requests to be aligned. This is needed
663 * so QEMU makes sure all IO operations on the device are aligned
664 * to sector size, or else FreeBSD will reject them with EINVAL.
666 s
->needs_alignment
= true;
671 if (platform_test_xfs_fd(s
->fd
)) {
676 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
679 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
686 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
689 BDRVRawState
*s
= bs
->opaque
;
691 s
->type
= FTYPE_FILE
;
692 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
701 #define PERM_FOREACH(i) \
702 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
704 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
705 * file; if @unlock == true, also unlock the unneeded bytes.
706 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
708 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
709 uint64_t perm_lock_bits
,
710 uint64_t shared_perm_lock_bits
,
711 bool unlock
, Error
**errp
)
715 uint64_t locked_perm
, locked_shared_perm
;
718 locked_perm
= s
->locked_perm
;
719 locked_shared_perm
= s
->locked_shared_perm
;
722 * We don't have the previous bits, just lock/unlock for each of the
726 locked_perm
= BLK_PERM_ALL
;
727 locked_shared_perm
= BLK_PERM_ALL
;
730 locked_shared_perm
= 0;
735 int off
= RAW_LOCK_PERM_BASE
+ i
;
736 uint64_t bit
= (1ULL << i
);
737 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
738 ret
= qemu_lock_fd(fd
, off
, 1, false);
740 error_setg(errp
, "Failed to lock byte %d", off
);
743 s
->locked_perm
|= bit
;
745 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
746 ret
= qemu_unlock_fd(fd
, off
, 1);
748 error_setg(errp
, "Failed to unlock byte %d", off
);
751 s
->locked_perm
&= ~bit
;
756 int off
= RAW_LOCK_SHARED_BASE
+ i
;
757 uint64_t bit
= (1ULL << i
);
758 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
759 ret
= qemu_lock_fd(fd
, off
, 1, false);
761 error_setg(errp
, "Failed to lock byte %d", off
);
764 s
->locked_shared_perm
|= bit
;
766 } else if (unlock
&& (locked_shared_perm
& bit
) &&
767 !(shared_perm_lock_bits
& bit
)) {
768 ret
= qemu_unlock_fd(fd
, off
, 1);
770 error_setg(errp
, "Failed to unlock byte %d", off
);
773 s
->locked_shared_perm
&= ~bit
;
780 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
781 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
788 int off
= RAW_LOCK_SHARED_BASE
+ i
;
789 uint64_t p
= 1ULL << i
;
791 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
793 char *perm_name
= bdrv_perm_names(p
);
795 "Failed to get \"%s\" lock",
803 int off
= RAW_LOCK_PERM_BASE
+ i
;
804 uint64_t p
= 1ULL << i
;
805 if (!(shared_perm
& p
)) {
806 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
808 char *perm_name
= bdrv_perm_names(p
);
810 "Failed to get shared \"%s\" lock",
820 static int raw_handle_perm_lock(BlockDriverState
*bs
,
822 uint64_t new_perm
, uint64_t new_shared
,
825 BDRVRawState
*s
= bs
->opaque
;
827 Error
*local_err
= NULL
;
833 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
839 if ((s
->perm
| new_perm
) == s
->perm
&&
840 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
843 * We are going to unlock bytes, it should not fail. If it fail due
844 * to some fs-dependent permission-unrelated reasons (which occurs
845 * sometimes on NFS and leads to abort in bdrv_replace_child) we
846 * can't prevent such errors by any check here. And we ignore them
847 * anyway in ABORT and COMMIT.
851 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
852 ~s
->shared_perm
| ~new_shared
,
855 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
859 error_append_hint(errp
,
860 "Is another process using the image [%s]?\n",
864 /* fall through to unlock bytes. */
866 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
869 /* Theoretically the above call only unlocks bytes and it cannot
870 * fail. Something weird happened, report it.
872 warn_report_err(local_err
);
876 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
879 /* Theoretically the above call only unlocks bytes and it cannot
880 * fail. Something weird happened, report it.
882 warn_report_err(local_err
);
889 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
890 int *open_flags
, uint64_t perm
, bool force_dup
,
893 BDRVRawState
*s
= bs
->opaque
;
896 bool has_writers
= perm
&
897 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
898 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
900 fcntl_flags
|= O_NOATIME
;
904 if (s
->type
== FTYPE_CD
) {
905 *open_flags
|= O_NONBLOCK
;
908 raw_parse_flags(flags
, open_flags
, has_writers
);
911 /* Not all operating systems have O_ASYNC, and those that don't
912 * will not let us track the state into rs->open_flags (typically
913 * you achieve the same effect with an ioctl, for example I_SETSIG
914 * on Solaris). But we do not use O_ASYNC, so that's fine.
916 assert((s
->open_flags
& O_ASYNC
) == 0);
919 if (!force_dup
&& *open_flags
== s
->open_flags
) {
920 /* We're lucky, the existing fd is fine */
924 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
925 /* dup the original fd */
926 fd
= qemu_dup(s
->fd
);
928 ret
= fcntl_setfl(fd
, *open_flags
);
936 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
938 const char *normalized_filename
= bs
->filename
;
939 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
941 assert(!(*open_flags
& O_CREAT
));
942 fd
= qemu_open(normalized_filename
, *open_flags
);
944 error_setg_errno(errp
, errno
, "Could not reopen file");
953 static int raw_reopen_prepare(BDRVReopenState
*state
,
954 BlockReopenQueue
*queue
, Error
**errp
)
957 BDRVRawReopenState
*rs
;
960 Error
*local_err
= NULL
;
962 assert(state
!= NULL
);
963 assert(state
->bs
!= NULL
);
965 s
= state
->bs
->opaque
;
967 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
970 /* Handle options changes */
971 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
972 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
974 error_propagate(errp
, local_err
);
979 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
980 rs
->check_cache_dropped
=
981 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
983 /* This driver's reopen function doesn't currently allow changing
984 * other options, so let's put them back in the original QDict and
985 * bdrv_reopen_prepare() will detect changes and complain. */
986 qemu_opts_to_qdict(opts
, state
->options
);
988 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
989 state
->perm
, true, &local_err
);
991 error_propagate(errp
, local_err
);
996 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
997 * alignment with the new fd. */
999 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
1001 error_propagate(errp
, local_err
);
1007 s
->reopen_state
= state
;
1015 qemu_opts_del(opts
);
1019 static void raw_reopen_commit(BDRVReopenState
*state
)
1021 BDRVRawReopenState
*rs
= state
->opaque
;
1022 BDRVRawState
*s
= state
->bs
->opaque
;
1024 s
->drop_cache
= rs
->drop_cache
;
1025 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1026 s
->open_flags
= rs
->open_flags
;
1031 g_free(state
->opaque
);
1032 state
->opaque
= NULL
;
1034 assert(s
->reopen_state
== state
);
1035 s
->reopen_state
= NULL
;
1039 static void raw_reopen_abort(BDRVReopenState
*state
)
1041 BDRVRawReopenState
*rs
= state
->opaque
;
1042 BDRVRawState
*s
= state
->bs
->opaque
;
1044 /* nothing to do if NULL, we didn't get far enough */
1053 g_free(state
->opaque
);
1054 state
->opaque
= NULL
;
1056 assert(s
->reopen_state
== state
);
1057 s
->reopen_state
= NULL
;
1060 static int sg_get_max_transfer_length(int fd
)
1065 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1075 static int sg_get_max_segments(int fd
)
1080 char *sysfspath
= NULL
;
1086 if (fstat(fd
, &st
)) {
1091 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1092 major(st
.st_rdev
), minor(st
.st_rdev
));
1093 sysfd
= open(sysfspath
, O_RDONLY
);
1099 ret
= read(sysfd
, buf
, sizeof(buf
) - 1);
1100 } while (ret
== -1 && errno
== EINTR
);
1104 } else if (ret
== 0) {
1109 /* The file is ended with '\n', pass 'end' to accept that. */
1110 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1111 if (ret
== 0 && end
&& *end
== '\n') {
1126 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1128 BDRVRawState
*s
= bs
->opaque
;
1131 int ret
= sg_get_max_transfer_length(s
->fd
);
1133 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1134 bs
->bl
.max_transfer
= pow2floor(ret
);
1137 ret
= sg_get_max_segments(s
->fd
);
1139 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
, ret
* getpagesize());
1143 raw_probe_alignment(bs
, s
->fd
, errp
);
1144 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1145 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, getpagesize());
1148 static int check_for_dasd(int fd
)
1151 struct dasd_information2_t info
= {0};
1153 return ioctl(fd
, BIODASDINFO2
, &info
);
1160 * Try to get @bs's logical and physical block size.
1161 * On success, store them in @bsz and return zero.
1162 * On failure, return negative errno.
1164 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1166 BDRVRawState
*s
= bs
->opaque
;
1169 /* If DASD, get blocksizes */
1170 if (check_for_dasd(s
->fd
) < 0) {
1173 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1177 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1181 * Try to get @bs's geometry: cyls, heads, sectors.
1182 * On success, store them in @geo and return 0.
1183 * On failure return -errno.
1184 * (Allows block driver to assign default geometry values that guest sees)
1187 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1189 BDRVRawState
*s
= bs
->opaque
;
1190 struct hd_geometry ioctl_geo
= {0};
1192 /* If DASD, get its geometry */
1193 if (check_for_dasd(s
->fd
) < 0) {
1196 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1199 /* HDIO_GETGEO may return success even though geo contains zeros
1200 (e.g. certain multipath setups) */
1201 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1204 /* Do not return a geometry for partition */
1205 if (ioctl_geo
.start
!= 0) {
1208 geo
->heads
= ioctl_geo
.heads
;
1209 geo
->sectors
= ioctl_geo
.sectors
;
1210 geo
->cylinders
= ioctl_geo
.cylinders
;
1214 #else /* __linux__ */
1215 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1221 #if defined(__linux__)
1222 static int handle_aiocb_ioctl(void *opaque
)
1224 RawPosixAIOData
*aiocb
= opaque
;
1227 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1236 static int handle_aiocb_flush(void *opaque
)
1238 RawPosixAIOData
*aiocb
= opaque
;
1239 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1242 if (s
->page_cache_inconsistent
) {
1246 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1248 /* There is no clear definition of the semantics of a failing fsync(),
1249 * so we may have to assume the worst. The sad truth is that this
1250 * assumption is correct for Linux. Some pages are now probably marked
1251 * clean in the page cache even though they are inconsistent with the
1252 * on-disk contents. The next fdatasync() call would succeed, but no
1253 * further writeback attempt will be made. We can't get back to a state
1254 * in which we know what is on disk (we would have to rewrite
1255 * everything that was touched since the last fdatasync() at least), so
1256 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1257 * really defined, I have little hope that other OSes are doing better.
1259 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1261 if ((s
->open_flags
& O_DIRECT
) == 0) {
1262 s
->page_cache_inconsistent
= true;
1269 #ifdef CONFIG_PREADV
1271 static bool preadv_present
= true;
1274 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1276 return preadv(fd
, iov
, nr_iov
, offset
);
1280 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1282 return pwritev(fd
, iov
, nr_iov
, offset
);
1287 static bool preadv_present
= false;
1290 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1296 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1303 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1308 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1309 len
= qemu_pwritev(aiocb
->aio_fildes
,
1314 len
= qemu_preadv(aiocb
->aio_fildes
,
1318 } while (len
== -1 && errno
== EINTR
);
1327 * Read/writes the data to/from a given linear buffer.
1329 * Returns the number of bytes handles or -errno in case of an error. Short
1330 * reads are only returned if the end of the file is reached.
1332 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1337 while (offset
< aiocb
->aio_nbytes
) {
1338 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1339 len
= pwrite(aiocb
->aio_fildes
,
1340 (const char *)buf
+ offset
,
1341 aiocb
->aio_nbytes
- offset
,
1342 aiocb
->aio_offset
+ offset
);
1344 len
= pread(aiocb
->aio_fildes
,
1346 aiocb
->aio_nbytes
- offset
,
1347 aiocb
->aio_offset
+ offset
);
1349 if (len
== -1 && errno
== EINTR
) {
1351 } else if (len
== -1 && errno
== EINVAL
&&
1352 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1353 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1355 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1356 * after a short read. Assume that O_DIRECT short reads only occur
1357 * at EOF. Therefore this is a short read, not an I/O error.
1360 } else if (len
== -1) {
1363 } else if (len
== 0) {
1372 static int handle_aiocb_rw(void *opaque
)
1374 RawPosixAIOData
*aiocb
= opaque
;
1378 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1380 * If there is just a single buffer, and it is properly aligned
1381 * we can just use plain pread/pwrite without any problems.
1383 if (aiocb
->io
.niov
== 1) {
1384 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1388 * We have more than one iovec, and all are properly aligned.
1390 * Try preadv/pwritev first and fall back to linearizing the
1391 * buffer if it's not supported.
1393 if (preadv_present
) {
1394 nbytes
= handle_aiocb_rw_vector(aiocb
);
1395 if (nbytes
== aiocb
->aio_nbytes
||
1396 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1399 preadv_present
= false;
1403 * XXX(hch): short read/write. no easy way to handle the reminder
1404 * using these interfaces. For now retry using plain
1410 * Ok, we have to do it the hard way, copy all segments into
1411 * a single aligned buffer.
1413 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1419 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1423 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1424 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1425 p
+= aiocb
->io
.iov
[i
].iov_len
;
1427 assert(p
- buf
== aiocb
->aio_nbytes
);
1430 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1431 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1433 size_t count
= aiocb
->aio_nbytes
, copy
;
1436 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1438 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1439 copy
= aiocb
->io
.iov
[i
].iov_len
;
1441 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1442 assert(count
>= copy
);
1451 if (nbytes
== aiocb
->aio_nbytes
) {
1453 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1454 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1457 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1458 0, aiocb
->aio_nbytes
- nbytes
);
1467 static int translate_err(int err
)
1469 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1476 #ifdef CONFIG_FALLOCATE
1477 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1480 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1483 } while (errno
== EINTR
);
1484 return translate_err(-errno
);
1488 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1491 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1493 if (!s
->has_write_zeroes
) {
1498 /* The BLKZEROOUT implementation in the kernel doesn't set
1499 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1501 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1503 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1504 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1507 } while (errno
== EINTR
);
1509 ret
= translate_err(-errno
);
1510 if (ret
== -ENOTSUP
) {
1511 s
->has_write_zeroes
= false;
1519 static int handle_aiocb_write_zeroes(void *opaque
)
1521 RawPosixAIOData
*aiocb
= opaque
;
1522 #ifdef CONFIG_FALLOCATE
1523 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1527 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1528 return handle_aiocb_write_zeroes_block(aiocb
);
1531 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1532 if (s
->has_write_zeroes
) {
1533 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1534 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1535 if (ret
== -EINVAL
) {
1537 * Allow falling back to pwrite for file systems that
1538 * do not support fallocate() for an unaligned byte range.
1542 if (ret
== 0 || ret
!= -ENOTSUP
) {
1545 s
->has_write_zeroes
= false;
1549 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1550 if (s
->has_discard
&& s
->has_fallocate
) {
1551 int ret
= do_fallocate(s
->fd
,
1552 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1553 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1555 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1556 if (ret
== 0 || ret
!= -ENOTSUP
) {
1559 s
->has_fallocate
= false;
1560 } else if (ret
!= -ENOTSUP
) {
1563 s
->has_discard
= false;
1568 #ifdef CONFIG_FALLOCATE
1569 /* Last resort: we are trying to extend the file with zeroed data. This
1570 * can be done via fallocate(fd, 0) */
1571 len
= bdrv_getlength(aiocb
->bs
);
1572 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1573 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1574 if (ret
== 0 || ret
!= -ENOTSUP
) {
1577 s
->has_fallocate
= false;
1584 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1586 RawPosixAIOData
*aiocb
= opaque
;
1587 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1590 /* First try to write zeros and unmap at the same time */
1592 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1593 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1594 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1595 if (ret
!= -ENOTSUP
) {
1600 /* If we couldn't manage to unmap while guaranteed that the area reads as
1601 * all-zero afterwards, just write zeroes without unmapping */
1602 ret
= handle_aiocb_write_zeroes(aiocb
);
1606 #ifndef HAVE_COPY_FILE_RANGE
1607 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1608 off_t
*out_off
, size_t len
, unsigned int flags
)
1610 #ifdef __NR_copy_file_range
1611 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1612 out_off
, len
, flags
);
1620 static int handle_aiocb_copy_range(void *opaque
)
1622 RawPosixAIOData
*aiocb
= opaque
;
1623 uint64_t bytes
= aiocb
->aio_nbytes
;
1624 off_t in_off
= aiocb
->aio_offset
;
1625 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1628 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1629 aiocb
->copy_range
.aio_fd2
, &out_off
,
1631 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1632 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1635 /* No progress (e.g. when beyond EOF), let the caller fall back to
1654 static int handle_aiocb_discard(void *opaque
)
1656 RawPosixAIOData
*aiocb
= opaque
;
1657 int ret
= -EOPNOTSUPP
;
1658 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1660 if (!s
->has_discard
) {
1664 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1667 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1668 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1671 } while (errno
== EINTR
);
1676 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1677 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1678 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1682 ret
= translate_err(ret
);
1683 if (ret
== -ENOTSUP
) {
1684 s
->has_discard
= false;
1690 * Help alignment probing by allocating the first block.
1692 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1693 * reading succeeds regardless of request length. In this case we fallback to
1694 * safe alignment which is not optimal. Allocating the first block avoids this
1697 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1698 * request alignment, so we use safe values.
1700 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1701 * caller may ignore failures.
1703 static int allocate_first_block(int fd
, size_t max_size
)
1705 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
1708 size_t max_align
= MAX(MAX_BLOCKSIZE
, getpagesize());
1713 buf
= qemu_memalign(max_align
, write_size
);
1714 memset(buf
, 0, write_size
);
1717 n
= pwrite(fd
, buf
, write_size
, 0);
1718 } while (n
== -1 && errno
== EINTR
);
1720 ret
= (n
== -1) ? -errno
: 0;
1726 static int handle_aiocb_truncate(void *opaque
)
1728 RawPosixAIOData
*aiocb
= opaque
;
1730 int64_t current_length
= 0;
1733 int fd
= aiocb
->aio_fildes
;
1734 int64_t offset
= aiocb
->aio_offset
;
1735 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1736 Error
**errp
= aiocb
->truncate
.errp
;
1738 if (fstat(fd
, &st
) < 0) {
1740 error_setg_errno(errp
, -result
, "Could not stat file");
1744 current_length
= st
.st_size
;
1745 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1746 error_setg(errp
, "Cannot use preallocation for shrinking files");
1751 #ifdef CONFIG_POSIX_FALLOCATE
1752 case PREALLOC_MODE_FALLOC
:
1754 * Truncating before posix_fallocate() makes it about twice slower on
1755 * file systems that do not support fallocate(), trying to check if a
1756 * block is allocated before allocating it, so don't do that here.
1758 if (offset
!= current_length
) {
1759 result
= -posix_fallocate(fd
, current_length
,
1760 offset
- current_length
);
1762 /* posix_fallocate() doesn't set errno. */
1763 error_setg_errno(errp
, -result
,
1764 "Could not preallocate new data");
1765 } else if (current_length
== 0) {
1767 * posix_fallocate() uses fallocate() if the filesystem
1768 * supports it, or fallback to manually writing zeroes. If
1769 * fallocate() was used, unaligned reads from the fallocated
1770 * area in raw_probe_alignment() will succeed, hence we need to
1771 * allocate the first block.
1773 * Optimize future alignment probing; ignore failures.
1775 allocate_first_block(fd
, offset
);
1782 case PREALLOC_MODE_FULL
:
1784 int64_t num
= 0, left
= offset
- current_length
;
1788 * Knowing the final size from the beginning could allow the file
1789 * system driver to do less allocations and possibly avoid
1790 * fragmentation of the file.
1792 if (ftruncate(fd
, offset
) != 0) {
1794 error_setg_errno(errp
, -result
, "Could not resize file");
1798 buf
= g_malloc0(65536);
1800 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1801 if (seek_result
< 0) {
1803 error_setg_errno(errp
, -result
,
1804 "Failed to seek to the old end of file");
1809 num
= MIN(left
, 65536);
1810 result
= write(fd
, buf
, num
);
1812 if (errno
== EINTR
) {
1816 error_setg_errno(errp
, -result
,
1817 "Could not write zeros for preallocation");
1826 error_setg_errno(errp
, -result
,
1827 "Could not flush file to disk");
1833 case PREALLOC_MODE_OFF
:
1834 if (ftruncate(fd
, offset
) != 0) {
1836 error_setg_errno(errp
, -result
, "Could not resize file");
1837 } else if (current_length
== 0 && offset
> current_length
) {
1838 /* Optimize future alignment probing; ignore failures. */
1839 allocate_first_block(fd
, offset
);
1844 error_setg(errp
, "Unsupported preallocation mode: %s",
1845 PreallocMode_str(prealloc
));
1851 if (ftruncate(fd
, current_length
) < 0) {
1852 error_report("Failed to restore old file length: %s",
1861 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1862 ThreadPoolFunc func
, void *arg
)
1864 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1865 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1866 return thread_pool_submit_co(pool
, func
, arg
);
1869 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1870 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1872 BDRVRawState
*s
= bs
->opaque
;
1873 RawPosixAIOData acb
;
1875 if (fd_open(bs
) < 0)
1879 * Check if the underlying device requires requests to be aligned,
1880 * and if the request we are trying to submit is aligned or not.
1881 * If this is the case tell the low-level driver that it needs
1882 * to copy the buffer.
1884 if (s
->needs_alignment
) {
1885 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1886 type
|= QEMU_AIO_MISALIGNED
;
1887 #ifdef CONFIG_LINUX_AIO
1888 } else if (s
->use_linux_aio
) {
1889 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1890 assert(qiov
->size
== bytes
);
1891 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1896 acb
= (RawPosixAIOData
) {
1898 .aio_fildes
= s
->fd
,
1900 .aio_offset
= offset
,
1901 .aio_nbytes
= bytes
,
1908 assert(qiov
->size
== bytes
);
1909 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1912 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1913 uint64_t bytes
, QEMUIOVector
*qiov
,
1916 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1919 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1920 uint64_t bytes
, QEMUIOVector
*qiov
,
1924 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1927 static void raw_aio_plug(BlockDriverState
*bs
)
1929 #ifdef CONFIG_LINUX_AIO
1930 BDRVRawState
*s
= bs
->opaque
;
1931 if (s
->use_linux_aio
) {
1932 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1933 laio_io_plug(bs
, aio
);
1938 static void raw_aio_unplug(BlockDriverState
*bs
)
1940 #ifdef CONFIG_LINUX_AIO
1941 BDRVRawState
*s
= bs
->opaque
;
1942 if (s
->use_linux_aio
) {
1943 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1944 laio_io_unplug(bs
, aio
);
1949 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1951 BDRVRawState
*s
= bs
->opaque
;
1952 RawPosixAIOData acb
;
1960 acb
= (RawPosixAIOData
) {
1962 .aio_fildes
= s
->fd
,
1963 .aio_type
= QEMU_AIO_FLUSH
,
1966 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
1969 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1970 AioContext
*new_context
)
1972 #ifdef CONFIG_LINUX_AIO
1973 BDRVRawState
*s
= bs
->opaque
;
1974 if (s
->use_linux_aio
) {
1976 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1977 error_reportf_err(local_err
, "Unable to use native AIO, "
1978 "falling back to thread pool: ");
1979 s
->use_linux_aio
= false;
1985 static void raw_close(BlockDriverState
*bs
)
1987 BDRVRawState
*s
= bs
->opaque
;
1996 * Truncates the given regular file @fd to @offset and, when growing, fills the
1997 * new space according to @prealloc.
1999 * Returns: 0 on success, -errno on failure.
2001 static int coroutine_fn
2002 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2003 PreallocMode prealloc
, Error
**errp
)
2005 RawPosixAIOData acb
;
2007 acb
= (RawPosixAIOData
) {
2010 .aio_type
= QEMU_AIO_TRUNCATE
,
2011 .aio_offset
= offset
,
2013 .prealloc
= prealloc
,
2018 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2021 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2022 bool exact
, PreallocMode prealloc
,
2025 BDRVRawState
*s
= bs
->opaque
;
2029 if (fstat(s
->fd
, &st
)) {
2031 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2035 if (S_ISREG(st
.st_mode
)) {
2036 /* Always resizes to the exact @offset */
2037 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2040 if (prealloc
!= PREALLOC_MODE_OFF
) {
2041 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2042 "non-regular file", PreallocMode_str(prealloc
));
2046 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2047 int64_t cur_length
= raw_getlength(bs
);
2049 if (offset
!= cur_length
&& exact
) {
2050 error_setg(errp
, "Cannot resize device files");
2052 } else if (offset
> cur_length
) {
2053 error_setg(errp
, "Cannot grow device files");
2057 error_setg(errp
, "Resizing this file is not supported");
2065 static int64_t raw_getlength(BlockDriverState
*bs
)
2067 BDRVRawState
*s
= bs
->opaque
;
2073 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2074 struct disklabel dl
;
2076 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2078 return (uint64_t)dl
.d_secsize
*
2079 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2083 #elif defined(__NetBSD__)
2084 static int64_t raw_getlength(BlockDriverState
*bs
)
2086 BDRVRawState
*s
= bs
->opaque
;
2092 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2093 struct dkwedge_info dkw
;
2095 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2096 return dkw
.dkw_size
* 512;
2098 struct disklabel dl
;
2100 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2102 return (uint64_t)dl
.d_secsize
*
2103 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2108 #elif defined(__sun__)
2109 static int64_t raw_getlength(BlockDriverState
*bs
)
2111 BDRVRawState
*s
= bs
->opaque
;
2112 struct dk_minfo minfo
;
2122 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2124 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2126 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2130 * There are reports that lseek on some devices fails, but
2131 * irc discussion said that contingency on contingency was overkill.
2133 size
= lseek(s
->fd
, 0, SEEK_END
);
2139 #elif defined(CONFIG_BSD)
2140 static int64_t raw_getlength(BlockDriverState
*bs
)
2142 BDRVRawState
*s
= bs
->opaque
;
2146 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2155 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2158 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2159 #ifdef DIOCGMEDIASIZE
2160 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2161 #elif defined(DIOCGPART)
2164 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2165 size
= pi
.media_size
;
2171 #if defined(__APPLE__) && defined(__MACH__)
2173 uint64_t sectors
= 0;
2174 uint32_t sector_size
= 0;
2176 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2177 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2178 size
= sectors
* sector_size
;
2180 size
= lseek(fd
, 0LL, SEEK_END
);
2187 size
= lseek(fd
, 0LL, SEEK_END
);
2192 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2195 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2196 if (size
== 2048LL * (unsigned)-1)
2198 /* XXX no disc? maybe we need to reopen... */
2199 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2206 size
= lseek(fd
, 0, SEEK_END
);
2214 static int64_t raw_getlength(BlockDriverState
*bs
)
2216 BDRVRawState
*s
= bs
->opaque
;
2225 size
= lseek(s
->fd
, 0, SEEK_END
);
2233 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2236 BDRVRawState
*s
= bs
->opaque
;
2238 if (fstat(s
->fd
, &st
) < 0) {
2241 return (int64_t)st
.st_blocks
* 512;
2244 static int coroutine_fn
2245 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2247 BlockdevCreateOptionsFile
*file_opts
;
2248 Error
*local_err
= NULL
;
2250 uint64_t perm
, shared
;
2253 /* Validate options and set default values */
2254 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2255 file_opts
= &options
->u
.file
;
2257 if (!file_opts
->has_nocow
) {
2258 file_opts
->nocow
= false;
2260 if (!file_opts
->has_preallocation
) {
2261 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2265 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2268 error_setg_errno(errp
, -result
, "Could not create file");
2272 /* Take permissions: We want to discard everything, so we need
2273 * BLK_PERM_WRITE; and truncation to the desired size requires
2275 * On the other hand, we cannot share the RESIZE permission
2276 * because we promise that after this function, the file has the
2277 * size given in the options. If someone else were to resize it
2278 * concurrently, we could not guarantee that.
2279 * Note that after this function, we can no longer guarantee that
2280 * the file is not touched by a third party, so it may be resized
2282 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2283 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2285 /* Step one: Take locks */
2286 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2291 /* Step two: Check that nobody else has taken conflicting locks */
2292 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2294 error_append_hint(errp
,
2295 "Is another process using the image [%s]?\n",
2296 file_opts
->filename
);
2300 /* Clear the file by truncating it to 0 */
2301 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2306 if (file_opts
->nocow
) {
2308 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2309 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2310 * will be ignored since any failure of this operation should not
2311 * block the left work.
2314 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2315 attr
|= FS_NOCOW_FL
;
2316 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2321 /* Resize and potentially preallocate the file to the desired
2323 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2324 file_opts
->preallocation
, errp
);
2330 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2332 /* The above call should not fail, and if it does, that does
2333 * not mean the whole creation operation has failed. So
2334 * report it the user for their convenience, but do not report
2335 * it to the caller. */
2336 warn_report_err(local_err
);
2340 if (qemu_close(fd
) != 0 && result
== 0) {
2342 error_setg_errno(errp
, -result
, "Could not close the new file");
2348 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2351 BlockdevCreateOptions options
;
2352 int64_t total_size
= 0;
2354 PreallocMode prealloc
;
2356 Error
*local_err
= NULL
;
2358 /* Skip file: protocol prefix */
2359 strstart(filename
, "file:", &filename
);
2361 /* Read out options */
2362 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2364 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2365 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2366 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2367 PREALLOC_MODE_OFF
, &local_err
);
2370 error_propagate(errp
, local_err
);
2374 options
= (BlockdevCreateOptions
) {
2375 .driver
= BLOCKDEV_DRIVER_FILE
,
2377 .filename
= (char *) filename
,
2379 .has_preallocation
= true,
2380 .preallocation
= prealloc
,
2385 return raw_co_create(&options
, errp
);
2389 * Find allocation range in @bs around offset @start.
2390 * May change underlying file descriptor's file offset.
2391 * If @start is not in a hole, store @start in @data, and the
2392 * beginning of the next hole in @hole, and return 0.
2393 * If @start is in a non-trailing hole, store @start in @hole and the
2394 * beginning of the next non-hole in @data, and return 0.
2395 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2396 * If we can't find out, return a negative errno other than -ENXIO.
2398 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2399 off_t
*data
, off_t
*hole
)
2401 #if defined SEEK_HOLE && defined SEEK_DATA
2402 BDRVRawState
*s
= bs
->opaque
;
2407 * D1. offs == start: start is in data
2408 * D2. offs > start: start is in a hole, next data at offs
2409 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2410 * or start is beyond EOF
2411 * If the latter happens, the file has been truncated behind
2412 * our back since we opened it. All bets are off then.
2413 * Treating like a trailing hole is simplest.
2414 * D4. offs < 0, errno != ENXIO: we learned nothing
2416 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2418 return -errno
; /* D3 or D4 */
2422 /* This is not a valid return by lseek(). We are safe to just return
2423 * -EIO in this case, and we'll treat it like D4. */
2428 /* D2: in hole, next data at offs */
2434 /* D1: in data, end not yet known */
2438 * H1. offs == start: start is in a hole
2439 * If this happens here, a hole has been dug behind our back
2440 * since the previous lseek().
2441 * H2. offs > start: either start is in data, next hole at offs,
2442 * or start is in trailing hole, EOF at offs
2443 * Linux treats trailing holes like any other hole: offs ==
2444 * start. Solaris seeks to EOF instead: offs > start (blech).
2445 * If that happens here, a hole has been dug behind our back
2446 * since the previous lseek().
2447 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2448 * If this happens, the file has been truncated behind our
2449 * back since we opened it. Treat it like a trailing hole.
2450 * H4. offs < 0, errno != ENXIO: we learned nothing
2451 * Pretend we know nothing at all, i.e. "forget" about D1.
2453 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2455 return -errno
; /* D1 and (H3 or H4) */
2459 /* This is not a valid return by lseek(). We are safe to just return
2460 * -EIO in this case, and we'll treat it like H4. */
2466 * D1 and H2: either in data, next hole at offs, or it was in
2467 * data but is now in a trailing hole. In the latter case,
2468 * all bets are off. Treating it as if it there was data all
2469 * the way to EOF is safe, so simply do that.
2484 * Returns the allocation status of the specified offset.
2486 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2488 * 'pnum' is set to the number of bytes (including and immediately following
2489 * the specified offset) that are known to be in the same
2490 * allocated/unallocated state.
2492 * 'bytes' is the max value 'pnum' should be set to.
2494 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2497 int64_t bytes
, int64_t *pnum
,
2499 BlockDriverState
**file
)
2501 off_t data
= 0, hole
= 0;
2504 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2515 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2518 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2519 if (ret
== -ENXIO
) {
2522 ret
= BDRV_BLOCK_ZERO
;
2523 } else if (ret
< 0) {
2524 /* No info available, so pretend there are no holes */
2526 ret
= BDRV_BLOCK_DATA
;
2527 } else if (data
== offset
) {
2528 /* On a data extent, compute bytes to the end of the extent,
2529 * possibly including a partial sector at EOF. */
2530 *pnum
= MIN(bytes
, hole
- offset
);
2533 * We are not allowed to return partial sectors, though, so
2534 * round up if necessary.
2536 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2537 int64_t file_length
= raw_getlength(bs
);
2538 if (file_length
> 0) {
2539 /* Ignore errors, this is just a safeguard */
2540 assert(hole
== file_length
);
2542 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2545 ret
= BDRV_BLOCK_DATA
;
2547 /* On a hole, compute bytes to the beginning of the next extent. */
2548 assert(hole
== offset
);
2549 *pnum
= MIN(bytes
, data
- offset
);
2550 ret
= BDRV_BLOCK_ZERO
;
2554 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2557 #if defined(__linux__)
2558 /* Verify that the file is not in the page cache */
2559 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2561 const size_t window_size
= 128 * 1024 * 1024;
2562 BDRVRawState
*s
= bs
->opaque
;
2563 void *window
= NULL
;
2570 /* mincore(2) page status information requires 1 byte per page */
2571 page_size
= sysconf(_SC_PAGESIZE
);
2572 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2574 end
= raw_getlength(bs
);
2576 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2583 /* Unmap previous window if size has changed */
2584 new_length
= MIN(end
- offset
, window_size
);
2585 if (new_length
!= length
) {
2586 munmap(window
, length
);
2591 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2593 if (new_window
== MAP_FAILED
) {
2594 error_setg_errno(errp
, errno
, "mmap failed");
2598 window
= new_window
;
2599 length
= new_length
;
2601 ret
= mincore(window
, length
, vec
);
2603 error_setg_errno(errp
, errno
, "mincore failed");
2607 vec_end
= DIV_ROUND_UP(length
, page_size
);
2608 for (i
= 0; i
< vec_end
; i
++) {
2610 error_setg(errp
, "page cache still in use!");
2617 munmap(window
, length
);
2622 #endif /* __linux__ */
2624 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2627 BDRVRawState
*s
= bs
->opaque
;
2632 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2636 if (!s
->drop_cache
) {
2640 if (s
->open_flags
& O_DIRECT
) {
2641 return; /* No host kernel page cache */
2644 #if defined(__linux__)
2645 /* This sets the scene for the next syscall... */
2646 ret
= bdrv_co_flush(bs
);
2648 error_setg_errno(errp
, -ret
, "flush failed");
2652 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2653 * process. These limitations are okay because we just fsynced the file,
2654 * we don't use mmap, and the file should not be in use by other processes.
2656 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2657 if (ret
!= 0) { /* the return value is a positive errno */
2658 error_setg_errno(errp
, ret
, "fadvise failed");
2662 if (s
->check_cache_dropped
) {
2663 check_cache_dropped(bs
, errp
);
2665 #else /* __linux__ */
2666 /* Do nothing. Live migration to a remote host with cache.direct=off is
2667 * unsupported on other host operating systems. Cache consistency issues
2668 * may occur but no error is reported here, partly because that's the
2669 * historical behavior and partly because it's hard to differentiate valid
2670 * configurations that should not cause errors.
2672 #endif /* !__linux__ */
2675 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
2678 s
->stats
.discard_nb_failed
++;
2680 s
->stats
.discard_nb_ok
++;
2681 s
->stats
.discard_bytes_ok
+= nbytes
;
2685 static coroutine_fn
int
2686 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2688 BDRVRawState
*s
= bs
->opaque
;
2689 RawPosixAIOData acb
;
2692 acb
= (RawPosixAIOData
) {
2694 .aio_fildes
= s
->fd
,
2695 .aio_type
= QEMU_AIO_DISCARD
,
2696 .aio_offset
= offset
,
2697 .aio_nbytes
= bytes
,
2701 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2704 ret
= raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2705 raw_account_discard(s
, bytes
, ret
);
2709 static coroutine_fn
int
2710 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2712 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2715 static int coroutine_fn
2716 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2717 BdrvRequestFlags flags
, bool blkdev
)
2719 BDRVRawState
*s
= bs
->opaque
;
2720 RawPosixAIOData acb
;
2721 ThreadPoolFunc
*handler
;
2723 acb
= (RawPosixAIOData
) {
2725 .aio_fildes
= s
->fd
,
2726 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2727 .aio_offset
= offset
,
2728 .aio_nbytes
= bytes
,
2732 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2734 if (flags
& BDRV_REQ_NO_FALLBACK
) {
2735 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
2738 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2739 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2740 handler
= handle_aiocb_write_zeroes_unmap
;
2742 handler
= handle_aiocb_write_zeroes
;
2745 return raw_thread_pool_submit(bs
, handler
, &acb
);
2748 static int coroutine_fn
raw_co_pwrite_zeroes(
2749 BlockDriverState
*bs
, int64_t offset
,
2750 int bytes
, BdrvRequestFlags flags
)
2752 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2755 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2757 BDRVRawState
*s
= bs
->opaque
;
2759 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2763 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
2765 BDRVRawState
*s
= bs
->opaque
;
2766 return (BlockStatsSpecificFile
) {
2767 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
2768 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
2769 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
2773 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
2775 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2777 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
2778 stats
->u
.file
= get_blockstats_specific_file(bs
);
2783 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
2785 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2787 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
2788 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
2793 static QemuOptsList raw_create_opts
= {
2794 .name
= "raw-create-opts",
2795 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2798 .name
= BLOCK_OPT_SIZE
,
2799 .type
= QEMU_OPT_SIZE
,
2800 .help
= "Virtual disk size"
2803 .name
= BLOCK_OPT_NOCOW
,
2804 .type
= QEMU_OPT_BOOL
,
2805 .help
= "Turn off copy-on-write (valid only on btrfs)"
2808 .name
= BLOCK_OPT_PREALLOC
,
2809 .type
= QEMU_OPT_STRING
,
2810 .help
= "Preallocation mode (allowed values: off"
2811 #ifdef CONFIG_POSIX_FALLOCATE
2816 { /* end of list */ }
2820 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2823 BDRVRawState
*s
= bs
->opaque
;
2824 BDRVRawReopenState
*rs
= NULL
;
2828 if (s
->perm_change_fd
) {
2830 * In the context of reopen, this function may be called several times
2831 * (directly and recursively while change permissions of the parent).
2832 * This is even true for children that don't inherit from the original
2833 * reopen node, so s->reopen_state is not set.
2835 * Ignore all but the first call.
2840 if (s
->reopen_state
) {
2841 /* We already have a new file descriptor to set permissions for */
2842 assert(s
->reopen_state
->perm
== perm
);
2843 assert(s
->reopen_state
->shared_perm
== shared
);
2844 rs
= s
->reopen_state
->opaque
;
2845 s
->perm_change_fd
= rs
->fd
;
2846 s
->perm_change_flags
= rs
->open_flags
;
2848 /* We may need a new fd if auto-read-only switches the mode */
2849 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
2853 } else if (ret
!= s
->fd
) {
2854 s
->perm_change_fd
= ret
;
2855 s
->perm_change_flags
= open_flags
;
2859 /* Prepare permissions on old fd to avoid conflicts between old and new,
2860 * but keep everything locked that new will need. */
2861 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2866 /* Copy locks to the new fd */
2867 if (s
->perm_change_fd
) {
2868 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
2871 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2878 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2879 qemu_close(s
->perm_change_fd
);
2881 s
->perm_change_fd
= 0;
2885 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2887 BDRVRawState
*s
= bs
->opaque
;
2889 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
2890 * called after .bdrv_reopen_commit) */
2891 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
2893 s
->fd
= s
->perm_change_fd
;
2894 s
->open_flags
= s
->perm_change_flags
;
2896 s
->perm_change_fd
= 0;
2898 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2900 s
->shared_perm
= shared
;
2903 static void raw_abort_perm_update(BlockDriverState
*bs
)
2905 BDRVRawState
*s
= bs
->opaque
;
2907 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
2908 * the file descriptor. */
2909 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2910 qemu_close(s
->perm_change_fd
);
2912 s
->perm_change_fd
= 0;
2914 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2917 static int coroutine_fn
raw_co_copy_range_from(
2918 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2919 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2920 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2922 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2923 read_flags
, write_flags
);
2926 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2928 uint64_t src_offset
,
2930 uint64_t dst_offset
,
2932 BdrvRequestFlags read_flags
,
2933 BdrvRequestFlags write_flags
)
2935 RawPosixAIOData acb
;
2936 BDRVRawState
*s
= bs
->opaque
;
2937 BDRVRawState
*src_s
;
2939 assert(dst
->bs
== bs
);
2940 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2944 src_s
= src
->bs
->opaque
;
2945 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2949 acb
= (RawPosixAIOData
) {
2951 .aio_type
= QEMU_AIO_COPY_RANGE
,
2952 .aio_fildes
= src_s
->fd
,
2953 .aio_offset
= src_offset
,
2954 .aio_nbytes
= bytes
,
2957 .aio_offset2
= dst_offset
,
2961 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
2964 BlockDriver bdrv_file
= {
2965 .format_name
= "file",
2966 .protocol_name
= "file",
2967 .instance_size
= sizeof(BDRVRawState
),
2968 .bdrv_needs_filename
= true,
2969 .bdrv_probe
= NULL
, /* no probe for protocols */
2970 .bdrv_parse_filename
= raw_parse_filename
,
2971 .bdrv_file_open
= raw_open
,
2972 .bdrv_reopen_prepare
= raw_reopen_prepare
,
2973 .bdrv_reopen_commit
= raw_reopen_commit
,
2974 .bdrv_reopen_abort
= raw_reopen_abort
,
2975 .bdrv_close
= raw_close
,
2976 .bdrv_co_create
= raw_co_create
,
2977 .bdrv_co_create_opts
= raw_co_create_opts
,
2978 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
2979 .bdrv_has_zero_init_truncate
= bdrv_has_zero_init_1
,
2980 .bdrv_co_block_status
= raw_co_block_status
,
2981 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
2982 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
2984 .bdrv_co_preadv
= raw_co_preadv
,
2985 .bdrv_co_pwritev
= raw_co_pwritev
,
2986 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
2987 .bdrv_co_pdiscard
= raw_co_pdiscard
,
2988 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
2989 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
2990 .bdrv_refresh_limits
= raw_refresh_limits
,
2991 .bdrv_io_plug
= raw_aio_plug
,
2992 .bdrv_io_unplug
= raw_aio_unplug
,
2993 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
2995 .bdrv_co_truncate
= raw_co_truncate
,
2996 .bdrv_getlength
= raw_getlength
,
2997 .bdrv_get_info
= raw_get_info
,
2998 .bdrv_get_allocated_file_size
2999 = raw_get_allocated_file_size
,
3000 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3001 .bdrv_check_perm
= raw_check_perm
,
3002 .bdrv_set_perm
= raw_set_perm
,
3003 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3004 .create_opts
= &raw_create_opts
,
3005 .mutable_opts
= mutable_opts
,
3008 /***********************************************/
3011 #if defined(__APPLE__) && defined(__MACH__)
3012 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3013 CFIndex maxPathSize
, int flags
);
3014 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3016 kern_return_t kernResult
= KERN_FAILURE
;
3017 mach_port_t masterPort
;
3018 CFMutableDictionaryRef classesToMatch
;
3019 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3020 char *mediaType
= NULL
;
3022 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
3023 if ( KERN_SUCCESS
!= kernResult
) {
3024 printf( "IOMasterPort returned %d\n", kernResult
);
3028 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3029 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3030 if (classesToMatch
== NULL
) {
3031 error_report("IOServiceMatching returned NULL for %s",
3032 matching_array
[index
]);
3035 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3037 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
3039 if (kernResult
!= KERN_SUCCESS
) {
3040 error_report("Note: IOServiceGetMatchingServices returned %d",
3045 /* If a match was found, leave the loop */
3046 if (*mediaIterator
!= 0) {
3047 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3048 mediaType
= g_strdup(matching_array
[index
]);
3055 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3056 CFIndex maxPathSize
, int flags
)
3058 io_object_t nextMedia
;
3059 kern_return_t kernResult
= KERN_FAILURE
;
3061 nextMedia
= IOIteratorNext( mediaIterator
);
3064 CFTypeRef bsdPathAsCFString
;
3065 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3066 if ( bsdPathAsCFString
) {
3067 size_t devPathLength
;
3068 strcpy( bsdPath
, _PATH_DEV
);
3069 if (flags
& BDRV_O_NOCACHE
) {
3070 strcat(bsdPath
, "r");
3072 devPathLength
= strlen( bsdPath
);
3073 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3074 kernResult
= KERN_SUCCESS
;
3076 CFRelease( bsdPathAsCFString
);
3078 IOObjectRelease( nextMedia
);
3084 /* Sets up a real cdrom for use in QEMU */
3085 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3087 int index
, num_of_test_partitions
= 2, fd
;
3088 char test_partition
[MAXPATHLEN
];
3089 bool partition_found
= false;
3091 /* look for a working partition */
3092 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3093 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3095 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
3097 partition_found
= true;
3103 /* if a working partition on the device was not found */
3104 if (partition_found
== false) {
3105 error_setg(errp
, "Failed to find a working partition on disc");
3107 trace_file_setup_cdrom(test_partition
);
3108 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3110 return partition_found
;
3113 /* Prints directions on mounting and unmounting a device */
3114 static void print_unmounting_directions(const char *file_name
)
3116 error_report("If device %s is mounted on the desktop, unmount"
3117 " it first before using it in QEMU", file_name
);
3118 error_report("Command to unmount device: diskutil unmountDisk %s",
3120 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3123 #endif /* defined(__APPLE__) && defined(__MACH__) */
3125 static int hdev_probe_device(const char *filename
)
3129 /* allow a dedicated CD-ROM driver to match with a higher priority */
3130 if (strstart(filename
, "/dev/cdrom", NULL
))
3133 if (stat(filename
, &st
) >= 0 &&
3134 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3141 static int check_hdev_writable(BDRVRawState
*s
)
3143 #if defined(BLKROGET)
3144 /* Linux block devices can be configured "read-only" using blockdev(8).
3145 * This is independent of device node permissions and therefore open(2)
3146 * with O_RDWR succeeds. Actual writes fail with EPERM.
3148 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3149 * check for read-only block devices so that Linux block devices behave
3155 if (fstat(s
->fd
, &st
)) {
3159 if (!S_ISBLK(st
.st_mode
)) {
3163 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
3170 #endif /* defined(BLKROGET) */
3174 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3177 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3180 static bool hdev_is_sg(BlockDriverState
*bs
)
3183 #if defined(__linux__)
3185 BDRVRawState
*s
= bs
->opaque
;
3187 struct sg_scsi_id scsiid
;
3191 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3195 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3200 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3202 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3211 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3214 BDRVRawState
*s
= bs
->opaque
;
3215 Error
*local_err
= NULL
;
3218 #if defined(__APPLE__) && defined(__MACH__)
3220 * Caution: while qdict_get_str() is fine, getting non-string types
3221 * would require more care. When @options come from -blockdev or
3222 * blockdev_add, its members are typed according to the QAPI
3223 * schema, but when they come from -drive, they're all QString.
3225 const char *filename
= qdict_get_str(options
, "filename");
3226 char bsd_path
[MAXPATHLEN
] = "";
3227 bool error_occurred
= false;
3229 /* If using a real cdrom */
3230 if (strcmp(filename
, "/dev/cdrom") == 0) {
3231 char *mediaType
= NULL
;
3232 kern_return_t ret_val
;
3233 io_iterator_t mediaIterator
= 0;
3235 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3236 if (mediaType
== NULL
) {
3237 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3239 error_occurred
= true;
3240 goto hdev_open_Mac_error
;
3243 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3244 if (ret_val
!= KERN_SUCCESS
) {
3245 error_setg(errp
, "Could not get BSD path for optical drive");
3246 error_occurred
= true;
3247 goto hdev_open_Mac_error
;
3250 /* If a real optical drive was not found */
3251 if (bsd_path
[0] == '\0') {
3252 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3253 error_occurred
= true;
3254 goto hdev_open_Mac_error
;
3257 /* If using a cdrom disc and finding a partition on the disc failed */
3258 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3259 setup_cdrom(bsd_path
, errp
) == false) {
3260 print_unmounting_directions(bsd_path
);
3261 error_occurred
= true;
3262 goto hdev_open_Mac_error
;
3265 qdict_put_str(options
, "filename", bsd_path
);
3267 hdev_open_Mac_error
:
3269 if (mediaIterator
) {
3270 IOObjectRelease(mediaIterator
);
3272 if (error_occurred
) {
3276 #endif /* defined(__APPLE__) && defined(__MACH__) */
3278 s
->type
= FTYPE_FILE
;
3280 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3282 error_propagate(errp
, local_err
);
3283 #if defined(__APPLE__) && defined(__MACH__)
3285 filename
= bsd_path
;
3287 /* if a physical device experienced an error while being opened */
3288 if (strncmp(filename
, "/dev/", 5) == 0) {
3289 print_unmounting_directions(filename
);
3291 #endif /* defined(__APPLE__) && defined(__MACH__) */
3295 /* Since this does ioctl the device must be already opened */
3296 bs
->sg
= hdev_is_sg(bs
);
3298 if (flags
& BDRV_O_RDWR
) {
3299 ret
= check_hdev_writable(s
);
3302 error_setg_errno(errp
, -ret
, "The device is not writable");
3310 #if defined(__linux__)
3311 static int coroutine_fn
3312 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3314 BDRVRawState
*s
= bs
->opaque
;
3315 RawPosixAIOData acb
;
3323 if (req
== SG_IO
&& s
->pr_mgr
) {
3324 struct sg_io_hdr
*io_hdr
= buf
;
3325 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3326 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3327 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3332 acb
= (RawPosixAIOData
) {
3334 .aio_type
= QEMU_AIO_IOCTL
,
3335 .aio_fildes
= s
->fd
,
3343 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3347 static int fd_open(BlockDriverState
*bs
)
3349 BDRVRawState
*s
= bs
->opaque
;
3351 /* this is just to ensure s->fd is sane (its called by io ops) */
3357 static coroutine_fn
int
3358 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3360 BDRVRawState
*s
= bs
->opaque
;
3365 raw_account_discard(s
, bytes
, ret
);
3368 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3371 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3372 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3381 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3384 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3389 struct stat stat_buf
;
3390 int64_t total_size
= 0;
3393 /* This function is used by both protocol block drivers and therefore either
3394 * of these prefixes may be given.
3395 * The return value has to be stored somewhere, otherwise this is an error
3396 * due to -Werror=unused-value. */
3398 strstart(filename
, "host_device:", &filename
) ||
3399 strstart(filename
, "host_cdrom:" , &filename
);
3403 ret
= raw_normalize_devicepath(&filename
, errp
);
3408 /* Read out options */
3409 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3412 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3415 error_setg_errno(errp
, -ret
, "Could not open device");
3419 if (fstat(fd
, &stat_buf
) < 0) {
3421 error_setg_errno(errp
, -ret
, "Could not stat device");
3422 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3424 "The given file is neither a block nor a character device");
3426 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3427 error_setg(errp
, "Device is too small");
3431 if (!ret
&& total_size
) {
3432 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3433 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3434 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3437 ret
= qemu_write_full(fd
, buf
, zero_size
);
3438 ret
= ret
== zero_size
? 0 : -errno
;
3445 static BlockDriver bdrv_host_device
= {
3446 .format_name
= "host_device",
3447 .protocol_name
= "host_device",
3448 .instance_size
= sizeof(BDRVRawState
),
3449 .bdrv_needs_filename
= true,
3450 .bdrv_probe_device
= hdev_probe_device
,
3451 .bdrv_parse_filename
= hdev_parse_filename
,
3452 .bdrv_file_open
= hdev_open
,
3453 .bdrv_close
= raw_close
,
3454 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3455 .bdrv_reopen_commit
= raw_reopen_commit
,
3456 .bdrv_reopen_abort
= raw_reopen_abort
,
3457 .bdrv_co_create_opts
= hdev_co_create_opts
,
3458 .create_opts
= &raw_create_opts
,
3459 .mutable_opts
= mutable_opts
,
3460 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3461 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3463 .bdrv_co_preadv
= raw_co_preadv
,
3464 .bdrv_co_pwritev
= raw_co_pwritev
,
3465 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3466 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3467 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3468 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3469 .bdrv_refresh_limits
= raw_refresh_limits
,
3470 .bdrv_io_plug
= raw_aio_plug
,
3471 .bdrv_io_unplug
= raw_aio_unplug
,
3472 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3474 .bdrv_co_truncate
= raw_co_truncate
,
3475 .bdrv_getlength
= raw_getlength
,
3476 .bdrv_get_info
= raw_get_info
,
3477 .bdrv_get_allocated_file_size
3478 = raw_get_allocated_file_size
,
3479 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
3480 .bdrv_check_perm
= raw_check_perm
,
3481 .bdrv_set_perm
= raw_set_perm
,
3482 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3483 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3484 .bdrv_probe_geometry
= hdev_probe_geometry
,
3486 /* generic scsi device */
3488 .bdrv_co_ioctl
= hdev_co_ioctl
,
3492 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3493 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3496 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3501 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3504 BDRVRawState
*s
= bs
->opaque
;
3508 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3509 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3512 static int cdrom_probe_device(const char *filename
)
3518 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3522 ret
= fstat(fd
, &st
);
3523 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3527 /* Attempt to detect via a CDROM specific ioctl */
3528 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3538 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3540 BDRVRawState
*s
= bs
->opaque
;
3543 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3544 return ret
== CDS_DISC_OK
;
3547 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3549 BDRVRawState
*s
= bs
->opaque
;
3552 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3553 perror("CDROMEJECT");
3555 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3556 perror("CDROMEJECT");
3560 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3562 BDRVRawState
*s
= bs
->opaque
;
3564 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3566 * Note: an error can happen if the distribution automatically
3569 /* perror("CDROM_LOCKDOOR"); */
3573 static BlockDriver bdrv_host_cdrom
= {
3574 .format_name
= "host_cdrom",
3575 .protocol_name
= "host_cdrom",
3576 .instance_size
= sizeof(BDRVRawState
),
3577 .bdrv_needs_filename
= true,
3578 .bdrv_probe_device
= cdrom_probe_device
,
3579 .bdrv_parse_filename
= cdrom_parse_filename
,
3580 .bdrv_file_open
= cdrom_open
,
3581 .bdrv_close
= raw_close
,
3582 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3583 .bdrv_reopen_commit
= raw_reopen_commit
,
3584 .bdrv_reopen_abort
= raw_reopen_abort
,
3585 .bdrv_co_create_opts
= hdev_co_create_opts
,
3586 .create_opts
= &raw_create_opts
,
3587 .mutable_opts
= mutable_opts
,
3588 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3591 .bdrv_co_preadv
= raw_co_preadv
,
3592 .bdrv_co_pwritev
= raw_co_pwritev
,
3593 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3594 .bdrv_refresh_limits
= raw_refresh_limits
,
3595 .bdrv_io_plug
= raw_aio_plug
,
3596 .bdrv_io_unplug
= raw_aio_unplug
,
3597 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3599 .bdrv_co_truncate
= raw_co_truncate
,
3600 .bdrv_getlength
= raw_getlength
,
3601 .has_variable_length
= true,
3602 .bdrv_get_allocated_file_size
3603 = raw_get_allocated_file_size
,
3605 /* removable device support */
3606 .bdrv_is_inserted
= cdrom_is_inserted
,
3607 .bdrv_eject
= cdrom_eject
,
3608 .bdrv_lock_medium
= cdrom_lock_medium
,
3610 /* generic scsi device */
3611 .bdrv_co_ioctl
= hdev_co_ioctl
,
3613 #endif /* __linux__ */
3615 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3616 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3619 BDRVRawState
*s
= bs
->opaque
;
3620 Error
*local_err
= NULL
;
3625 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3627 error_propagate(errp
, local_err
);
3631 /* make sure the door isn't locked at this time */
3632 ioctl(s
->fd
, CDIOCALLOW
);
3636 static int cdrom_probe_device(const char *filename
)
3638 if (strstart(filename
, "/dev/cd", NULL
) ||
3639 strstart(filename
, "/dev/acd", NULL
))
3644 static int cdrom_reopen(BlockDriverState
*bs
)
3646 BDRVRawState
*s
= bs
->opaque
;
3650 * Force reread of possibly changed/newly loaded disc,
3651 * FreeBSD seems to not notice sometimes...
3655 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3662 /* make sure the door isn't locked at this time */
3663 ioctl(s
->fd
, CDIOCALLOW
);
3667 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3669 return raw_getlength(bs
) > 0;
3672 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3674 BDRVRawState
*s
= bs
->opaque
;
3679 (void) ioctl(s
->fd
, CDIOCALLOW
);
3682 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3683 perror("CDIOCEJECT");
3685 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3686 perror("CDIOCCLOSE");
3692 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3694 BDRVRawState
*s
= bs
->opaque
;
3698 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3700 * Note: an error can happen if the distribution automatically
3703 /* perror("CDROM_LOCKDOOR"); */
3707 static BlockDriver bdrv_host_cdrom
= {
3708 .format_name
= "host_cdrom",
3709 .protocol_name
= "host_cdrom",
3710 .instance_size
= sizeof(BDRVRawState
),
3711 .bdrv_needs_filename
= true,
3712 .bdrv_probe_device
= cdrom_probe_device
,
3713 .bdrv_parse_filename
= cdrom_parse_filename
,
3714 .bdrv_file_open
= cdrom_open
,
3715 .bdrv_close
= raw_close
,
3716 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3717 .bdrv_reopen_commit
= raw_reopen_commit
,
3718 .bdrv_reopen_abort
= raw_reopen_abort
,
3719 .bdrv_co_create_opts
= hdev_co_create_opts
,
3720 .create_opts
= &raw_create_opts
,
3721 .mutable_opts
= mutable_opts
,
3723 .bdrv_co_preadv
= raw_co_preadv
,
3724 .bdrv_co_pwritev
= raw_co_pwritev
,
3725 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3726 .bdrv_refresh_limits
= raw_refresh_limits
,
3727 .bdrv_io_plug
= raw_aio_plug
,
3728 .bdrv_io_unplug
= raw_aio_unplug
,
3729 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3731 .bdrv_co_truncate
= raw_co_truncate
,
3732 .bdrv_getlength
= raw_getlength
,
3733 .has_variable_length
= true,
3734 .bdrv_get_allocated_file_size
3735 = raw_get_allocated_file_size
,
3737 /* removable device support */
3738 .bdrv_is_inserted
= cdrom_is_inserted
,
3739 .bdrv_eject
= cdrom_eject
,
3740 .bdrv_lock_medium
= cdrom_lock_medium
,
3742 #endif /* __FreeBSD__ */
3744 static void bdrv_file_init(void)
3747 * Register all the drivers. Note that order is important, the driver
3748 * registered last will get probed first.
3750 bdrv_register(&bdrv_file
);
3751 bdrv_register(&bdrv_host_device
);
3753 bdrv_register(&bdrv_host_cdrom
);
3755 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3756 bdrv_register(&bdrv_host_cdrom
);
3760 block_init(bdrv_file_init
);