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
, qemu_real_host_page_size
);
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
,
1140 ret
* qemu_real_host_page_size
);
1144 raw_probe_alignment(bs
, s
->fd
, errp
);
1145 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1146 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, qemu_real_host_page_size
);
1149 static int check_for_dasd(int fd
)
1152 struct dasd_information2_t info
= {0};
1154 return ioctl(fd
, BIODASDINFO2
, &info
);
1161 * Try to get @bs's logical and physical block size.
1162 * On success, store them in @bsz and return zero.
1163 * On failure, return negative errno.
1165 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1167 BDRVRawState
*s
= bs
->opaque
;
1170 /* If DASD, get blocksizes */
1171 if (check_for_dasd(s
->fd
) < 0) {
1174 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1178 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1182 * Try to get @bs's geometry: cyls, heads, sectors.
1183 * On success, store them in @geo and return 0.
1184 * On failure return -errno.
1185 * (Allows block driver to assign default geometry values that guest sees)
1188 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1190 BDRVRawState
*s
= bs
->opaque
;
1191 struct hd_geometry ioctl_geo
= {0};
1193 /* If DASD, get its geometry */
1194 if (check_for_dasd(s
->fd
) < 0) {
1197 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1200 /* HDIO_GETGEO may return success even though geo contains zeros
1201 (e.g. certain multipath setups) */
1202 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1205 /* Do not return a geometry for partition */
1206 if (ioctl_geo
.start
!= 0) {
1209 geo
->heads
= ioctl_geo
.heads
;
1210 geo
->sectors
= ioctl_geo
.sectors
;
1211 geo
->cylinders
= ioctl_geo
.cylinders
;
1215 #else /* __linux__ */
1216 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1222 #if defined(__linux__)
1223 static int handle_aiocb_ioctl(void *opaque
)
1225 RawPosixAIOData
*aiocb
= opaque
;
1228 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1237 static int handle_aiocb_flush(void *opaque
)
1239 RawPosixAIOData
*aiocb
= opaque
;
1240 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1243 if (s
->page_cache_inconsistent
) {
1247 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1249 /* There is no clear definition of the semantics of a failing fsync(),
1250 * so we may have to assume the worst. The sad truth is that this
1251 * assumption is correct for Linux. Some pages are now probably marked
1252 * clean in the page cache even though they are inconsistent with the
1253 * on-disk contents. The next fdatasync() call would succeed, but no
1254 * further writeback attempt will be made. We can't get back to a state
1255 * in which we know what is on disk (we would have to rewrite
1256 * everything that was touched since the last fdatasync() at least), so
1257 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1258 * really defined, I have little hope that other OSes are doing better.
1260 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1262 if ((s
->open_flags
& O_DIRECT
) == 0) {
1263 s
->page_cache_inconsistent
= true;
1270 #ifdef CONFIG_PREADV
1272 static bool preadv_present
= true;
1275 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1277 return preadv(fd
, iov
, nr_iov
, offset
);
1281 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1283 return pwritev(fd
, iov
, nr_iov
, offset
);
1288 static bool preadv_present
= false;
1291 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1297 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1304 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1309 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1310 len
= qemu_pwritev(aiocb
->aio_fildes
,
1315 len
= qemu_preadv(aiocb
->aio_fildes
,
1319 } while (len
== -1 && errno
== EINTR
);
1328 * Read/writes the data to/from a given linear buffer.
1330 * Returns the number of bytes handles or -errno in case of an error. Short
1331 * reads are only returned if the end of the file is reached.
1333 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1338 while (offset
< aiocb
->aio_nbytes
) {
1339 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1340 len
= pwrite(aiocb
->aio_fildes
,
1341 (const char *)buf
+ offset
,
1342 aiocb
->aio_nbytes
- offset
,
1343 aiocb
->aio_offset
+ offset
);
1345 len
= pread(aiocb
->aio_fildes
,
1347 aiocb
->aio_nbytes
- offset
,
1348 aiocb
->aio_offset
+ offset
);
1350 if (len
== -1 && errno
== EINTR
) {
1352 } else if (len
== -1 && errno
== EINVAL
&&
1353 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1354 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1356 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1357 * after a short read. Assume that O_DIRECT short reads only occur
1358 * at EOF. Therefore this is a short read, not an I/O error.
1361 } else if (len
== -1) {
1364 } else if (len
== 0) {
1373 static int handle_aiocb_rw(void *opaque
)
1375 RawPosixAIOData
*aiocb
= opaque
;
1379 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1381 * If there is just a single buffer, and it is properly aligned
1382 * we can just use plain pread/pwrite without any problems.
1384 if (aiocb
->io
.niov
== 1) {
1385 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1389 * We have more than one iovec, and all are properly aligned.
1391 * Try preadv/pwritev first and fall back to linearizing the
1392 * buffer if it's not supported.
1394 if (preadv_present
) {
1395 nbytes
= handle_aiocb_rw_vector(aiocb
);
1396 if (nbytes
== aiocb
->aio_nbytes
||
1397 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1400 preadv_present
= false;
1404 * XXX(hch): short read/write. no easy way to handle the reminder
1405 * using these interfaces. For now retry using plain
1411 * Ok, we have to do it the hard way, copy all segments into
1412 * a single aligned buffer.
1414 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1420 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1424 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1425 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1426 p
+= aiocb
->io
.iov
[i
].iov_len
;
1428 assert(p
- buf
== aiocb
->aio_nbytes
);
1431 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1432 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1434 size_t count
= aiocb
->aio_nbytes
, copy
;
1437 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1439 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1440 copy
= aiocb
->io
.iov
[i
].iov_len
;
1442 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1443 assert(count
>= copy
);
1452 if (nbytes
== aiocb
->aio_nbytes
) {
1454 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1455 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1458 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1459 0, aiocb
->aio_nbytes
- nbytes
);
1468 static int translate_err(int err
)
1470 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1477 #ifdef CONFIG_FALLOCATE
1478 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1481 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1484 } while (errno
== EINTR
);
1485 return translate_err(-errno
);
1489 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1492 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1494 if (!s
->has_write_zeroes
) {
1499 /* The BLKZEROOUT implementation in the kernel doesn't set
1500 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1502 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1504 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1505 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1508 } while (errno
== EINTR
);
1510 ret
= translate_err(-errno
);
1511 if (ret
== -ENOTSUP
) {
1512 s
->has_write_zeroes
= false;
1520 static int handle_aiocb_write_zeroes(void *opaque
)
1522 RawPosixAIOData
*aiocb
= opaque
;
1523 #ifdef CONFIG_FALLOCATE
1524 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1528 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1529 return handle_aiocb_write_zeroes_block(aiocb
);
1532 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1533 if (s
->has_write_zeroes
) {
1534 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1535 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1536 if (ret
== -EINVAL
) {
1538 * Allow falling back to pwrite for file systems that
1539 * do not support fallocate() for an unaligned byte range.
1543 if (ret
== 0 || ret
!= -ENOTSUP
) {
1546 s
->has_write_zeroes
= false;
1550 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1551 if (s
->has_discard
&& s
->has_fallocate
) {
1552 int ret
= do_fallocate(s
->fd
,
1553 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1554 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1556 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1557 if (ret
== 0 || ret
!= -ENOTSUP
) {
1560 s
->has_fallocate
= false;
1561 } else if (ret
!= -ENOTSUP
) {
1564 s
->has_discard
= false;
1569 #ifdef CONFIG_FALLOCATE
1570 /* Last resort: we are trying to extend the file with zeroed data. This
1571 * can be done via fallocate(fd, 0) */
1572 len
= bdrv_getlength(aiocb
->bs
);
1573 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1574 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1575 if (ret
== 0 || ret
!= -ENOTSUP
) {
1578 s
->has_fallocate
= false;
1585 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1587 RawPosixAIOData
*aiocb
= opaque
;
1588 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1591 /* First try to write zeros and unmap at the same time */
1593 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1594 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1595 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1596 if (ret
!= -ENOTSUP
) {
1601 /* If we couldn't manage to unmap while guaranteed that the area reads as
1602 * all-zero afterwards, just write zeroes without unmapping */
1603 ret
= handle_aiocb_write_zeroes(aiocb
);
1607 #ifndef HAVE_COPY_FILE_RANGE
1608 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1609 off_t
*out_off
, size_t len
, unsigned int flags
)
1611 #ifdef __NR_copy_file_range
1612 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1613 out_off
, len
, flags
);
1621 static int handle_aiocb_copy_range(void *opaque
)
1623 RawPosixAIOData
*aiocb
= opaque
;
1624 uint64_t bytes
= aiocb
->aio_nbytes
;
1625 off_t in_off
= aiocb
->aio_offset
;
1626 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1629 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1630 aiocb
->copy_range
.aio_fd2
, &out_off
,
1632 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1633 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1636 /* No progress (e.g. when beyond EOF), let the caller fall back to
1655 static int handle_aiocb_discard(void *opaque
)
1657 RawPosixAIOData
*aiocb
= opaque
;
1658 int ret
= -EOPNOTSUPP
;
1659 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1661 if (!s
->has_discard
) {
1665 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1668 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1669 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1672 } while (errno
== EINTR
);
1677 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1678 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1679 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1683 ret
= translate_err(ret
);
1684 if (ret
== -ENOTSUP
) {
1685 s
->has_discard
= false;
1691 * Help alignment probing by allocating the first block.
1693 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1694 * reading succeeds regardless of request length. In this case we fallback to
1695 * safe alignment which is not optimal. Allocating the first block avoids this
1698 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1699 * request alignment, so we use safe values.
1701 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1702 * caller may ignore failures.
1704 static int allocate_first_block(int fd
, size_t max_size
)
1706 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
1709 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size
);
1714 buf
= qemu_memalign(max_align
, write_size
);
1715 memset(buf
, 0, write_size
);
1718 n
= pwrite(fd
, buf
, write_size
, 0);
1719 } while (n
== -1 && errno
== EINTR
);
1721 ret
= (n
== -1) ? -errno
: 0;
1727 static int handle_aiocb_truncate(void *opaque
)
1729 RawPosixAIOData
*aiocb
= opaque
;
1731 int64_t current_length
= 0;
1734 int fd
= aiocb
->aio_fildes
;
1735 int64_t offset
= aiocb
->aio_offset
;
1736 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1737 Error
**errp
= aiocb
->truncate
.errp
;
1739 if (fstat(fd
, &st
) < 0) {
1741 error_setg_errno(errp
, -result
, "Could not stat file");
1745 current_length
= st
.st_size
;
1746 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1747 error_setg(errp
, "Cannot use preallocation for shrinking files");
1752 #ifdef CONFIG_POSIX_FALLOCATE
1753 case PREALLOC_MODE_FALLOC
:
1755 * Truncating before posix_fallocate() makes it about twice slower on
1756 * file systems that do not support fallocate(), trying to check if a
1757 * block is allocated before allocating it, so don't do that here.
1759 if (offset
!= current_length
) {
1760 result
= -posix_fallocate(fd
, current_length
,
1761 offset
- current_length
);
1763 /* posix_fallocate() doesn't set errno. */
1764 error_setg_errno(errp
, -result
,
1765 "Could not preallocate new data");
1766 } else if (current_length
== 0) {
1768 * posix_fallocate() uses fallocate() if the filesystem
1769 * supports it, or fallback to manually writing zeroes. If
1770 * fallocate() was used, unaligned reads from the fallocated
1771 * area in raw_probe_alignment() will succeed, hence we need to
1772 * allocate the first block.
1774 * Optimize future alignment probing; ignore failures.
1776 allocate_first_block(fd
, offset
);
1783 case PREALLOC_MODE_FULL
:
1785 int64_t num
= 0, left
= offset
- current_length
;
1789 * Knowing the final size from the beginning could allow the file
1790 * system driver to do less allocations and possibly avoid
1791 * fragmentation of the file.
1793 if (ftruncate(fd
, offset
) != 0) {
1795 error_setg_errno(errp
, -result
, "Could not resize file");
1799 buf
= g_malloc0(65536);
1801 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1802 if (seek_result
< 0) {
1804 error_setg_errno(errp
, -result
,
1805 "Failed to seek to the old end of file");
1810 num
= MIN(left
, 65536);
1811 result
= write(fd
, buf
, num
);
1813 if (errno
== EINTR
) {
1817 error_setg_errno(errp
, -result
,
1818 "Could not write zeros for preallocation");
1827 error_setg_errno(errp
, -result
,
1828 "Could not flush file to disk");
1834 case PREALLOC_MODE_OFF
:
1835 if (ftruncate(fd
, offset
) != 0) {
1837 error_setg_errno(errp
, -result
, "Could not resize file");
1838 } else if (current_length
== 0 && offset
> current_length
) {
1839 /* Optimize future alignment probing; ignore failures. */
1840 allocate_first_block(fd
, offset
);
1845 error_setg(errp
, "Unsupported preallocation mode: %s",
1846 PreallocMode_str(prealloc
));
1852 if (ftruncate(fd
, current_length
) < 0) {
1853 error_report("Failed to restore old file length: %s",
1862 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1863 ThreadPoolFunc func
, void *arg
)
1865 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1866 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1867 return thread_pool_submit_co(pool
, func
, arg
);
1870 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1871 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1873 BDRVRawState
*s
= bs
->opaque
;
1874 RawPosixAIOData acb
;
1876 if (fd_open(bs
) < 0)
1880 * Check if the underlying device requires requests to be aligned,
1881 * and if the request we are trying to submit is aligned or not.
1882 * If this is the case tell the low-level driver that it needs
1883 * to copy the buffer.
1885 if (s
->needs_alignment
) {
1886 if (!bdrv_qiov_is_aligned(bs
, qiov
)) {
1887 type
|= QEMU_AIO_MISALIGNED
;
1888 #ifdef CONFIG_LINUX_AIO
1889 } else if (s
->use_linux_aio
) {
1890 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1891 assert(qiov
->size
== bytes
);
1892 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1897 acb
= (RawPosixAIOData
) {
1899 .aio_fildes
= s
->fd
,
1901 .aio_offset
= offset
,
1902 .aio_nbytes
= bytes
,
1909 assert(qiov
->size
== bytes
);
1910 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1913 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1914 uint64_t bytes
, QEMUIOVector
*qiov
,
1917 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1920 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1921 uint64_t bytes
, QEMUIOVector
*qiov
,
1925 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1928 static void raw_aio_plug(BlockDriverState
*bs
)
1930 #ifdef CONFIG_LINUX_AIO
1931 BDRVRawState
*s
= bs
->opaque
;
1932 if (s
->use_linux_aio
) {
1933 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1934 laio_io_plug(bs
, aio
);
1939 static void raw_aio_unplug(BlockDriverState
*bs
)
1941 #ifdef CONFIG_LINUX_AIO
1942 BDRVRawState
*s
= bs
->opaque
;
1943 if (s
->use_linux_aio
) {
1944 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1945 laio_io_unplug(bs
, aio
);
1950 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1952 BDRVRawState
*s
= bs
->opaque
;
1953 RawPosixAIOData acb
;
1961 acb
= (RawPosixAIOData
) {
1963 .aio_fildes
= s
->fd
,
1964 .aio_type
= QEMU_AIO_FLUSH
,
1967 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
1970 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
1971 AioContext
*new_context
)
1973 #ifdef CONFIG_LINUX_AIO
1974 BDRVRawState
*s
= bs
->opaque
;
1975 if (s
->use_linux_aio
) {
1976 Error
*local_err
= NULL
;
1977 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
1978 error_reportf_err(local_err
, "Unable to use native AIO, "
1979 "falling back to thread pool: ");
1980 s
->use_linux_aio
= false;
1986 static void raw_close(BlockDriverState
*bs
)
1988 BDRVRawState
*s
= bs
->opaque
;
1997 * Truncates the given regular file @fd to @offset and, when growing, fills the
1998 * new space according to @prealloc.
2000 * Returns: 0 on success, -errno on failure.
2002 static int coroutine_fn
2003 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2004 PreallocMode prealloc
, Error
**errp
)
2006 RawPosixAIOData acb
;
2008 acb
= (RawPosixAIOData
) {
2011 .aio_type
= QEMU_AIO_TRUNCATE
,
2012 .aio_offset
= offset
,
2014 .prealloc
= prealloc
,
2019 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2022 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2023 bool exact
, PreallocMode prealloc
,
2026 BDRVRawState
*s
= bs
->opaque
;
2030 if (fstat(s
->fd
, &st
)) {
2032 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2036 if (S_ISREG(st
.st_mode
)) {
2037 /* Always resizes to the exact @offset */
2038 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2041 if (prealloc
!= PREALLOC_MODE_OFF
) {
2042 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2043 "non-regular file", PreallocMode_str(prealloc
));
2047 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2048 int64_t cur_length
= raw_getlength(bs
);
2050 if (offset
!= cur_length
&& exact
) {
2051 error_setg(errp
, "Cannot resize device files");
2053 } else if (offset
> cur_length
) {
2054 error_setg(errp
, "Cannot grow device files");
2058 error_setg(errp
, "Resizing this file is not supported");
2066 static int64_t raw_getlength(BlockDriverState
*bs
)
2068 BDRVRawState
*s
= bs
->opaque
;
2074 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2075 struct disklabel dl
;
2077 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2079 return (uint64_t)dl
.d_secsize
*
2080 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2084 #elif defined(__NetBSD__)
2085 static int64_t raw_getlength(BlockDriverState
*bs
)
2087 BDRVRawState
*s
= bs
->opaque
;
2093 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2094 struct dkwedge_info dkw
;
2096 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2097 return dkw
.dkw_size
* 512;
2099 struct disklabel dl
;
2101 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2103 return (uint64_t)dl
.d_secsize
*
2104 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2109 #elif defined(__sun__)
2110 static int64_t raw_getlength(BlockDriverState
*bs
)
2112 BDRVRawState
*s
= bs
->opaque
;
2113 struct dk_minfo minfo
;
2123 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2125 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2127 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2131 * There are reports that lseek on some devices fails, but
2132 * irc discussion said that contingency on contingency was overkill.
2134 size
= lseek(s
->fd
, 0, SEEK_END
);
2140 #elif defined(CONFIG_BSD)
2141 static int64_t raw_getlength(BlockDriverState
*bs
)
2143 BDRVRawState
*s
= bs
->opaque
;
2147 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2156 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2159 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2160 #ifdef DIOCGMEDIASIZE
2161 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2162 #elif defined(DIOCGPART)
2165 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2166 size
= pi
.media_size
;
2172 #if defined(__APPLE__) && defined(__MACH__)
2174 uint64_t sectors
= 0;
2175 uint32_t sector_size
= 0;
2177 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2178 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2179 size
= sectors
* sector_size
;
2181 size
= lseek(fd
, 0LL, SEEK_END
);
2188 size
= lseek(fd
, 0LL, SEEK_END
);
2193 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2196 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2197 if (size
== 2048LL * (unsigned)-1)
2199 /* XXX no disc? maybe we need to reopen... */
2200 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2207 size
= lseek(fd
, 0, SEEK_END
);
2215 static int64_t raw_getlength(BlockDriverState
*bs
)
2217 BDRVRawState
*s
= bs
->opaque
;
2226 size
= lseek(s
->fd
, 0, SEEK_END
);
2234 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2237 BDRVRawState
*s
= bs
->opaque
;
2239 if (fstat(s
->fd
, &st
) < 0) {
2242 return (int64_t)st
.st_blocks
* 512;
2245 static int coroutine_fn
2246 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2248 BlockdevCreateOptionsFile
*file_opts
;
2249 Error
*local_err
= NULL
;
2251 uint64_t perm
, shared
;
2254 /* Validate options and set default values */
2255 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2256 file_opts
= &options
->u
.file
;
2258 if (!file_opts
->has_nocow
) {
2259 file_opts
->nocow
= false;
2261 if (!file_opts
->has_preallocation
) {
2262 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2266 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2269 error_setg_errno(errp
, -result
, "Could not create file");
2273 /* Take permissions: We want to discard everything, so we need
2274 * BLK_PERM_WRITE; and truncation to the desired size requires
2276 * On the other hand, we cannot share the RESIZE permission
2277 * because we promise that after this function, the file has the
2278 * size given in the options. If someone else were to resize it
2279 * concurrently, we could not guarantee that.
2280 * Note that after this function, we can no longer guarantee that
2281 * the file is not touched by a third party, so it may be resized
2283 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2284 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2286 /* Step one: Take locks */
2287 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2292 /* Step two: Check that nobody else has taken conflicting locks */
2293 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2295 error_append_hint(errp
,
2296 "Is another process using the image [%s]?\n",
2297 file_opts
->filename
);
2301 /* Clear the file by truncating it to 0 */
2302 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2307 if (file_opts
->nocow
) {
2309 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2310 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2311 * will be ignored since any failure of this operation should not
2312 * block the left work.
2315 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2316 attr
|= FS_NOCOW_FL
;
2317 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2322 /* Resize and potentially preallocate the file to the desired
2324 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2325 file_opts
->preallocation
, errp
);
2331 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2333 /* The above call should not fail, and if it does, that does
2334 * not mean the whole creation operation has failed. So
2335 * report it the user for their convenience, but do not report
2336 * it to the caller. */
2337 warn_report_err(local_err
);
2341 if (qemu_close(fd
) != 0 && result
== 0) {
2343 error_setg_errno(errp
, -result
, "Could not close the new file");
2349 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2352 BlockdevCreateOptions options
;
2353 int64_t total_size
= 0;
2355 PreallocMode prealloc
;
2357 Error
*local_err
= NULL
;
2359 /* Skip file: protocol prefix */
2360 strstart(filename
, "file:", &filename
);
2362 /* Read out options */
2363 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2365 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2366 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2367 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2368 PREALLOC_MODE_OFF
, &local_err
);
2371 error_propagate(errp
, local_err
);
2375 options
= (BlockdevCreateOptions
) {
2376 .driver
= BLOCKDEV_DRIVER_FILE
,
2378 .filename
= (char *) filename
,
2380 .has_preallocation
= true,
2381 .preallocation
= prealloc
,
2386 return raw_co_create(&options
, errp
);
2390 * Find allocation range in @bs around offset @start.
2391 * May change underlying file descriptor's file offset.
2392 * If @start is not in a hole, store @start in @data, and the
2393 * beginning of the next hole in @hole, and return 0.
2394 * If @start is in a non-trailing hole, store @start in @hole and the
2395 * beginning of the next non-hole in @data, and return 0.
2396 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2397 * If we can't find out, return a negative errno other than -ENXIO.
2399 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2400 off_t
*data
, off_t
*hole
)
2402 #if defined SEEK_HOLE && defined SEEK_DATA
2403 BDRVRawState
*s
= bs
->opaque
;
2408 * D1. offs == start: start is in data
2409 * D2. offs > start: start is in a hole, next data at offs
2410 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2411 * or start is beyond EOF
2412 * If the latter happens, the file has been truncated behind
2413 * our back since we opened it. All bets are off then.
2414 * Treating like a trailing hole is simplest.
2415 * D4. offs < 0, errno != ENXIO: we learned nothing
2417 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2419 return -errno
; /* D3 or D4 */
2423 /* This is not a valid return by lseek(). We are safe to just return
2424 * -EIO in this case, and we'll treat it like D4. */
2429 /* D2: in hole, next data at offs */
2435 /* D1: in data, end not yet known */
2439 * H1. offs == start: start is in a hole
2440 * If this happens here, a hole has been dug behind our back
2441 * since the previous lseek().
2442 * H2. offs > start: either start is in data, next hole at offs,
2443 * or start is in trailing hole, EOF at offs
2444 * Linux treats trailing holes like any other hole: offs ==
2445 * start. Solaris seeks to EOF instead: offs > start (blech).
2446 * If that happens here, a hole has been dug behind our back
2447 * since the previous lseek().
2448 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2449 * If this happens, the file has been truncated behind our
2450 * back since we opened it. Treat it like a trailing hole.
2451 * H4. offs < 0, errno != ENXIO: we learned nothing
2452 * Pretend we know nothing at all, i.e. "forget" about D1.
2454 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2456 return -errno
; /* D1 and (H3 or H4) */
2460 /* This is not a valid return by lseek(). We are safe to just return
2461 * -EIO in this case, and we'll treat it like H4. */
2467 * D1 and H2: either in data, next hole at offs, or it was in
2468 * data but is now in a trailing hole. In the latter case,
2469 * all bets are off. Treating it as if it there was data all
2470 * the way to EOF is safe, so simply do that.
2485 * Returns the allocation status of the specified offset.
2487 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2489 * 'pnum' is set to the number of bytes (including and immediately following
2490 * the specified offset) that are known to be in the same
2491 * allocated/unallocated state.
2493 * 'bytes' is the max value 'pnum' should be set to.
2495 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2498 int64_t bytes
, int64_t *pnum
,
2500 BlockDriverState
**file
)
2502 off_t data
= 0, hole
= 0;
2505 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2516 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2519 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2520 if (ret
== -ENXIO
) {
2523 ret
= BDRV_BLOCK_ZERO
;
2524 } else if (ret
< 0) {
2525 /* No info available, so pretend there are no holes */
2527 ret
= BDRV_BLOCK_DATA
;
2528 } else if (data
== offset
) {
2529 /* On a data extent, compute bytes to the end of the extent,
2530 * possibly including a partial sector at EOF. */
2531 *pnum
= MIN(bytes
, hole
- offset
);
2534 * We are not allowed to return partial sectors, though, so
2535 * round up if necessary.
2537 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2538 int64_t file_length
= raw_getlength(bs
);
2539 if (file_length
> 0) {
2540 /* Ignore errors, this is just a safeguard */
2541 assert(hole
== file_length
);
2543 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2546 ret
= BDRV_BLOCK_DATA
;
2548 /* On a hole, compute bytes to the beginning of the next extent. */
2549 assert(hole
== offset
);
2550 *pnum
= MIN(bytes
, data
- offset
);
2551 ret
= BDRV_BLOCK_ZERO
;
2555 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2558 #if defined(__linux__)
2559 /* Verify that the file is not in the page cache */
2560 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2562 const size_t window_size
= 128 * 1024 * 1024;
2563 BDRVRawState
*s
= bs
->opaque
;
2564 void *window
= NULL
;
2571 /* mincore(2) page status information requires 1 byte per page */
2572 page_size
= sysconf(_SC_PAGESIZE
);
2573 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2575 end
= raw_getlength(bs
);
2577 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2584 /* Unmap previous window if size has changed */
2585 new_length
= MIN(end
- offset
, window_size
);
2586 if (new_length
!= length
) {
2587 munmap(window
, length
);
2592 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2594 if (new_window
== MAP_FAILED
) {
2595 error_setg_errno(errp
, errno
, "mmap failed");
2599 window
= new_window
;
2600 length
= new_length
;
2602 ret
= mincore(window
, length
, vec
);
2604 error_setg_errno(errp
, errno
, "mincore failed");
2608 vec_end
= DIV_ROUND_UP(length
, page_size
);
2609 for (i
= 0; i
< vec_end
; i
++) {
2611 error_setg(errp
, "page cache still in use!");
2618 munmap(window
, length
);
2623 #endif /* __linux__ */
2625 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2628 BDRVRawState
*s
= bs
->opaque
;
2633 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2637 if (!s
->drop_cache
) {
2641 if (s
->open_flags
& O_DIRECT
) {
2642 return; /* No host kernel page cache */
2645 #if defined(__linux__)
2646 /* This sets the scene for the next syscall... */
2647 ret
= bdrv_co_flush(bs
);
2649 error_setg_errno(errp
, -ret
, "flush failed");
2653 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2654 * process. These limitations are okay because we just fsynced the file,
2655 * we don't use mmap, and the file should not be in use by other processes.
2657 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2658 if (ret
!= 0) { /* the return value is a positive errno */
2659 error_setg_errno(errp
, ret
, "fadvise failed");
2663 if (s
->check_cache_dropped
) {
2664 check_cache_dropped(bs
, errp
);
2666 #else /* __linux__ */
2667 /* Do nothing. Live migration to a remote host with cache.direct=off is
2668 * unsupported on other host operating systems. Cache consistency issues
2669 * may occur but no error is reported here, partly because that's the
2670 * historical behavior and partly because it's hard to differentiate valid
2671 * configurations that should not cause errors.
2673 #endif /* !__linux__ */
2676 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
2679 s
->stats
.discard_nb_failed
++;
2681 s
->stats
.discard_nb_ok
++;
2682 s
->stats
.discard_bytes_ok
+= nbytes
;
2686 static coroutine_fn
int
2687 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2689 BDRVRawState
*s
= bs
->opaque
;
2690 RawPosixAIOData acb
;
2693 acb
= (RawPosixAIOData
) {
2695 .aio_fildes
= s
->fd
,
2696 .aio_type
= QEMU_AIO_DISCARD
,
2697 .aio_offset
= offset
,
2698 .aio_nbytes
= bytes
,
2702 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2705 ret
= raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2706 raw_account_discard(s
, bytes
, ret
);
2710 static coroutine_fn
int
2711 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2713 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2716 static int coroutine_fn
2717 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2718 BdrvRequestFlags flags
, bool blkdev
)
2720 BDRVRawState
*s
= bs
->opaque
;
2721 RawPosixAIOData acb
;
2722 ThreadPoolFunc
*handler
;
2724 #ifdef CONFIG_FALLOCATE
2725 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
2726 BdrvTrackedRequest
*req
;
2730 * This is a workaround for a bug in the Linux XFS driver,
2731 * where writes submitted through the AIO interface will be
2732 * discarded if they happen beyond a concurrently running
2733 * fallocate() that increases the file length (i.e., both the
2734 * write and the fallocate() happen beyond the EOF).
2736 * To work around it, we extend the tracked request for this
2737 * zero write until INT64_MAX (effectively infinity), and mark
2738 * it as serializing.
2740 * We have to enable this workaround for all filesystems and
2741 * AIO modes (not just XFS with aio=native), because for
2742 * remote filesystems we do not know the host configuration.
2745 req
= bdrv_co_get_self_request(bs
);
2747 assert(req
->type
== BDRV_TRACKED_WRITE
);
2748 assert(req
->offset
<= offset
);
2749 assert(req
->offset
+ req
->bytes
>= offset
+ bytes
);
2751 end
= INT64_MAX
& -(uint64_t)bs
->bl
.request_alignment
;
2752 req
->bytes
= end
- req
->offset
;
2753 req
->overlap_bytes
= req
->bytes
;
2755 bdrv_mark_request_serialising(req
, bs
->bl
.request_alignment
);
2756 bdrv_wait_serialising_requests(req
);
2760 acb
= (RawPosixAIOData
) {
2762 .aio_fildes
= s
->fd
,
2763 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2764 .aio_offset
= offset
,
2765 .aio_nbytes
= bytes
,
2769 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2771 if (flags
& BDRV_REQ_NO_FALLBACK
) {
2772 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
2775 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2776 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2777 handler
= handle_aiocb_write_zeroes_unmap
;
2779 handler
= handle_aiocb_write_zeroes
;
2782 return raw_thread_pool_submit(bs
, handler
, &acb
);
2785 static int coroutine_fn
raw_co_pwrite_zeroes(
2786 BlockDriverState
*bs
, int64_t offset
,
2787 int bytes
, BdrvRequestFlags flags
)
2789 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2792 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2794 BDRVRawState
*s
= bs
->opaque
;
2796 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2800 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
2802 BDRVRawState
*s
= bs
->opaque
;
2803 return (BlockStatsSpecificFile
) {
2804 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
2805 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
2806 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
2810 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
2812 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2814 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
2815 stats
->u
.file
= get_blockstats_specific_file(bs
);
2820 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
2822 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2824 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
2825 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
2830 static QemuOptsList raw_create_opts
= {
2831 .name
= "raw-create-opts",
2832 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2835 .name
= BLOCK_OPT_SIZE
,
2836 .type
= QEMU_OPT_SIZE
,
2837 .help
= "Virtual disk size"
2840 .name
= BLOCK_OPT_NOCOW
,
2841 .type
= QEMU_OPT_BOOL
,
2842 .help
= "Turn off copy-on-write (valid only on btrfs)"
2845 .name
= BLOCK_OPT_PREALLOC
,
2846 .type
= QEMU_OPT_STRING
,
2847 .help
= "Preallocation mode (allowed values: off"
2848 #ifdef CONFIG_POSIX_FALLOCATE
2853 { /* end of list */ }
2857 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2860 BDRVRawState
*s
= bs
->opaque
;
2861 BDRVRawReopenState
*rs
= NULL
;
2865 if (s
->perm_change_fd
) {
2867 * In the context of reopen, this function may be called several times
2868 * (directly and recursively while change permissions of the parent).
2869 * This is even true for children that don't inherit from the original
2870 * reopen node, so s->reopen_state is not set.
2872 * Ignore all but the first call.
2877 if (s
->reopen_state
) {
2878 /* We already have a new file descriptor to set permissions for */
2879 assert(s
->reopen_state
->perm
== perm
);
2880 assert(s
->reopen_state
->shared_perm
== shared
);
2881 rs
= s
->reopen_state
->opaque
;
2882 s
->perm_change_fd
= rs
->fd
;
2883 s
->perm_change_flags
= rs
->open_flags
;
2885 /* We may need a new fd if auto-read-only switches the mode */
2886 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
2890 } else if (ret
!= s
->fd
) {
2891 s
->perm_change_fd
= ret
;
2892 s
->perm_change_flags
= open_flags
;
2896 /* Prepare permissions on old fd to avoid conflicts between old and new,
2897 * but keep everything locked that new will need. */
2898 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2903 /* Copy locks to the new fd */
2904 if (s
->perm_change_fd
) {
2905 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
2908 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2915 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2916 qemu_close(s
->perm_change_fd
);
2918 s
->perm_change_fd
= 0;
2922 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2924 BDRVRawState
*s
= bs
->opaque
;
2926 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
2927 * called after .bdrv_reopen_commit) */
2928 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
2930 s
->fd
= s
->perm_change_fd
;
2931 s
->open_flags
= s
->perm_change_flags
;
2933 s
->perm_change_fd
= 0;
2935 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2937 s
->shared_perm
= shared
;
2940 static void raw_abort_perm_update(BlockDriverState
*bs
)
2942 BDRVRawState
*s
= bs
->opaque
;
2944 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
2945 * the file descriptor. */
2946 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2947 qemu_close(s
->perm_change_fd
);
2949 s
->perm_change_fd
= 0;
2951 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2954 static int coroutine_fn
raw_co_copy_range_from(
2955 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
2956 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
2957 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
2959 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2960 read_flags
, write_flags
);
2963 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
2965 uint64_t src_offset
,
2967 uint64_t dst_offset
,
2969 BdrvRequestFlags read_flags
,
2970 BdrvRequestFlags write_flags
)
2972 RawPosixAIOData acb
;
2973 BDRVRawState
*s
= bs
->opaque
;
2974 BDRVRawState
*src_s
;
2976 assert(dst
->bs
== bs
);
2977 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
2981 src_s
= src
->bs
->opaque
;
2982 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
2986 acb
= (RawPosixAIOData
) {
2988 .aio_type
= QEMU_AIO_COPY_RANGE
,
2989 .aio_fildes
= src_s
->fd
,
2990 .aio_offset
= src_offset
,
2991 .aio_nbytes
= bytes
,
2994 .aio_offset2
= dst_offset
,
2998 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
3001 BlockDriver bdrv_file
= {
3002 .format_name
= "file",
3003 .protocol_name
= "file",
3004 .instance_size
= sizeof(BDRVRawState
),
3005 .bdrv_needs_filename
= true,
3006 .bdrv_probe
= NULL
, /* no probe for protocols */
3007 .bdrv_parse_filename
= raw_parse_filename
,
3008 .bdrv_file_open
= raw_open
,
3009 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3010 .bdrv_reopen_commit
= raw_reopen_commit
,
3011 .bdrv_reopen_abort
= raw_reopen_abort
,
3012 .bdrv_close
= raw_close
,
3013 .bdrv_co_create
= raw_co_create
,
3014 .bdrv_co_create_opts
= raw_co_create_opts
,
3015 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3016 .bdrv_has_zero_init_truncate
= bdrv_has_zero_init_1
,
3017 .bdrv_co_block_status
= raw_co_block_status
,
3018 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3019 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
3021 .bdrv_co_preadv
= raw_co_preadv
,
3022 .bdrv_co_pwritev
= raw_co_pwritev
,
3023 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3024 .bdrv_co_pdiscard
= raw_co_pdiscard
,
3025 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3026 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3027 .bdrv_refresh_limits
= raw_refresh_limits
,
3028 .bdrv_io_plug
= raw_aio_plug
,
3029 .bdrv_io_unplug
= raw_aio_unplug
,
3030 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3032 .bdrv_co_truncate
= raw_co_truncate
,
3033 .bdrv_getlength
= raw_getlength
,
3034 .bdrv_get_info
= raw_get_info
,
3035 .bdrv_get_allocated_file_size
3036 = raw_get_allocated_file_size
,
3037 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3038 .bdrv_check_perm
= raw_check_perm
,
3039 .bdrv_set_perm
= raw_set_perm
,
3040 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3041 .create_opts
= &raw_create_opts
,
3042 .mutable_opts
= mutable_opts
,
3045 /***********************************************/
3048 #if defined(__APPLE__) && defined(__MACH__)
3049 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3050 CFIndex maxPathSize
, int flags
);
3051 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3053 kern_return_t kernResult
= KERN_FAILURE
;
3054 mach_port_t masterPort
;
3055 CFMutableDictionaryRef classesToMatch
;
3056 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3057 char *mediaType
= NULL
;
3059 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
3060 if ( KERN_SUCCESS
!= kernResult
) {
3061 printf( "IOMasterPort returned %d\n", kernResult
);
3065 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3066 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3067 if (classesToMatch
== NULL
) {
3068 error_report("IOServiceMatching returned NULL for %s",
3069 matching_array
[index
]);
3072 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3074 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
3076 if (kernResult
!= KERN_SUCCESS
) {
3077 error_report("Note: IOServiceGetMatchingServices returned %d",
3082 /* If a match was found, leave the loop */
3083 if (*mediaIterator
!= 0) {
3084 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3085 mediaType
= g_strdup(matching_array
[index
]);
3092 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3093 CFIndex maxPathSize
, int flags
)
3095 io_object_t nextMedia
;
3096 kern_return_t kernResult
= KERN_FAILURE
;
3098 nextMedia
= IOIteratorNext( mediaIterator
);
3101 CFTypeRef bsdPathAsCFString
;
3102 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3103 if ( bsdPathAsCFString
) {
3104 size_t devPathLength
;
3105 strcpy( bsdPath
, _PATH_DEV
);
3106 if (flags
& BDRV_O_NOCACHE
) {
3107 strcat(bsdPath
, "r");
3109 devPathLength
= strlen( bsdPath
);
3110 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3111 kernResult
= KERN_SUCCESS
;
3113 CFRelease( bsdPathAsCFString
);
3115 IOObjectRelease( nextMedia
);
3121 /* Sets up a real cdrom for use in QEMU */
3122 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3124 int index
, num_of_test_partitions
= 2, fd
;
3125 char test_partition
[MAXPATHLEN
];
3126 bool partition_found
= false;
3128 /* look for a working partition */
3129 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3130 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3132 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
3134 partition_found
= true;
3140 /* if a working partition on the device was not found */
3141 if (partition_found
== false) {
3142 error_setg(errp
, "Failed to find a working partition on disc");
3144 trace_file_setup_cdrom(test_partition
);
3145 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3147 return partition_found
;
3150 /* Prints directions on mounting and unmounting a device */
3151 static void print_unmounting_directions(const char *file_name
)
3153 error_report("If device %s is mounted on the desktop, unmount"
3154 " it first before using it in QEMU", file_name
);
3155 error_report("Command to unmount device: diskutil unmountDisk %s",
3157 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3160 #endif /* defined(__APPLE__) && defined(__MACH__) */
3162 static int hdev_probe_device(const char *filename
)
3166 /* allow a dedicated CD-ROM driver to match with a higher priority */
3167 if (strstart(filename
, "/dev/cdrom", NULL
))
3170 if (stat(filename
, &st
) >= 0 &&
3171 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3178 static int check_hdev_writable(BDRVRawState
*s
)
3180 #if defined(BLKROGET)
3181 /* Linux block devices can be configured "read-only" using blockdev(8).
3182 * This is independent of device node permissions and therefore open(2)
3183 * with O_RDWR succeeds. Actual writes fail with EPERM.
3185 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3186 * check for read-only block devices so that Linux block devices behave
3192 if (fstat(s
->fd
, &st
)) {
3196 if (!S_ISBLK(st
.st_mode
)) {
3200 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
3207 #endif /* defined(BLKROGET) */
3211 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3214 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3217 static bool hdev_is_sg(BlockDriverState
*bs
)
3220 #if defined(__linux__)
3222 BDRVRawState
*s
= bs
->opaque
;
3224 struct sg_scsi_id scsiid
;
3228 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3232 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3237 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3239 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3248 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3251 BDRVRawState
*s
= bs
->opaque
;
3252 Error
*local_err
= NULL
;
3255 #if defined(__APPLE__) && defined(__MACH__)
3257 * Caution: while qdict_get_str() is fine, getting non-string types
3258 * would require more care. When @options come from -blockdev or
3259 * blockdev_add, its members are typed according to the QAPI
3260 * schema, but when they come from -drive, they're all QString.
3262 const char *filename
= qdict_get_str(options
, "filename");
3263 char bsd_path
[MAXPATHLEN
] = "";
3264 bool error_occurred
= false;
3266 /* If using a real cdrom */
3267 if (strcmp(filename
, "/dev/cdrom") == 0) {
3268 char *mediaType
= NULL
;
3269 kern_return_t ret_val
;
3270 io_iterator_t mediaIterator
= 0;
3272 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3273 if (mediaType
== NULL
) {
3274 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3276 error_occurred
= true;
3277 goto hdev_open_Mac_error
;
3280 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3281 if (ret_val
!= KERN_SUCCESS
) {
3282 error_setg(errp
, "Could not get BSD path for optical drive");
3283 error_occurred
= true;
3284 goto hdev_open_Mac_error
;
3287 /* If a real optical drive was not found */
3288 if (bsd_path
[0] == '\0') {
3289 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3290 error_occurred
= true;
3291 goto hdev_open_Mac_error
;
3294 /* If using a cdrom disc and finding a partition on the disc failed */
3295 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3296 setup_cdrom(bsd_path
, errp
) == false) {
3297 print_unmounting_directions(bsd_path
);
3298 error_occurred
= true;
3299 goto hdev_open_Mac_error
;
3302 qdict_put_str(options
, "filename", bsd_path
);
3304 hdev_open_Mac_error
:
3306 if (mediaIterator
) {
3307 IOObjectRelease(mediaIterator
);
3309 if (error_occurred
) {
3313 #endif /* defined(__APPLE__) && defined(__MACH__) */
3315 s
->type
= FTYPE_FILE
;
3317 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3319 error_propagate(errp
, local_err
);
3320 #if defined(__APPLE__) && defined(__MACH__)
3322 filename
= bsd_path
;
3324 /* if a physical device experienced an error while being opened */
3325 if (strncmp(filename
, "/dev/", 5) == 0) {
3326 print_unmounting_directions(filename
);
3328 #endif /* defined(__APPLE__) && defined(__MACH__) */
3332 /* Since this does ioctl the device must be already opened */
3333 bs
->sg
= hdev_is_sg(bs
);
3335 if (flags
& BDRV_O_RDWR
) {
3336 ret
= check_hdev_writable(s
);
3339 error_setg_errno(errp
, -ret
, "The device is not writable");
3347 #if defined(__linux__)
3348 static int coroutine_fn
3349 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3351 BDRVRawState
*s
= bs
->opaque
;
3352 RawPosixAIOData acb
;
3360 if (req
== SG_IO
&& s
->pr_mgr
) {
3361 struct sg_io_hdr
*io_hdr
= buf
;
3362 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3363 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3364 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3369 acb
= (RawPosixAIOData
) {
3371 .aio_type
= QEMU_AIO_IOCTL
,
3372 .aio_fildes
= s
->fd
,
3380 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3384 static int fd_open(BlockDriverState
*bs
)
3386 BDRVRawState
*s
= bs
->opaque
;
3388 /* this is just to ensure s->fd is sane (its called by io ops) */
3394 static coroutine_fn
int
3395 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3397 BDRVRawState
*s
= bs
->opaque
;
3402 raw_account_discard(s
, bytes
, ret
);
3405 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3408 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3409 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3418 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3421 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3426 struct stat stat_buf
;
3427 int64_t total_size
= 0;
3430 /* This function is used by both protocol block drivers and therefore either
3431 * of these prefixes may be given.
3432 * The return value has to be stored somewhere, otherwise this is an error
3433 * due to -Werror=unused-value. */
3435 strstart(filename
, "host_device:", &filename
) ||
3436 strstart(filename
, "host_cdrom:" , &filename
);
3440 ret
= raw_normalize_devicepath(&filename
, errp
);
3445 /* Read out options */
3446 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3449 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3452 error_setg_errno(errp
, -ret
, "Could not open device");
3456 if (fstat(fd
, &stat_buf
) < 0) {
3458 error_setg_errno(errp
, -ret
, "Could not stat device");
3459 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3461 "The given file is neither a block nor a character device");
3463 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3464 error_setg(errp
, "Device is too small");
3468 if (!ret
&& total_size
) {
3469 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3470 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3471 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3474 ret
= qemu_write_full(fd
, buf
, zero_size
);
3475 ret
= ret
== zero_size
? 0 : -errno
;
3482 static BlockDriver bdrv_host_device
= {
3483 .format_name
= "host_device",
3484 .protocol_name
= "host_device",
3485 .instance_size
= sizeof(BDRVRawState
),
3486 .bdrv_needs_filename
= true,
3487 .bdrv_probe_device
= hdev_probe_device
,
3488 .bdrv_parse_filename
= hdev_parse_filename
,
3489 .bdrv_file_open
= hdev_open
,
3490 .bdrv_close
= raw_close
,
3491 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3492 .bdrv_reopen_commit
= raw_reopen_commit
,
3493 .bdrv_reopen_abort
= raw_reopen_abort
,
3494 .bdrv_co_create_opts
= hdev_co_create_opts
,
3495 .create_opts
= &raw_create_opts
,
3496 .mutable_opts
= mutable_opts
,
3497 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3498 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3500 .bdrv_co_preadv
= raw_co_preadv
,
3501 .bdrv_co_pwritev
= raw_co_pwritev
,
3502 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3503 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3504 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3505 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3506 .bdrv_refresh_limits
= raw_refresh_limits
,
3507 .bdrv_io_plug
= raw_aio_plug
,
3508 .bdrv_io_unplug
= raw_aio_unplug
,
3509 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3511 .bdrv_co_truncate
= raw_co_truncate
,
3512 .bdrv_getlength
= raw_getlength
,
3513 .bdrv_get_info
= raw_get_info
,
3514 .bdrv_get_allocated_file_size
3515 = raw_get_allocated_file_size
,
3516 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
3517 .bdrv_check_perm
= raw_check_perm
,
3518 .bdrv_set_perm
= raw_set_perm
,
3519 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3520 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3521 .bdrv_probe_geometry
= hdev_probe_geometry
,
3523 /* generic scsi device */
3525 .bdrv_co_ioctl
= hdev_co_ioctl
,
3529 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3530 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3533 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3538 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3541 BDRVRawState
*s
= bs
->opaque
;
3545 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3546 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3549 static int cdrom_probe_device(const char *filename
)
3555 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3559 ret
= fstat(fd
, &st
);
3560 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3564 /* Attempt to detect via a CDROM specific ioctl */
3565 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3575 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3577 BDRVRawState
*s
= bs
->opaque
;
3580 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3581 return ret
== CDS_DISC_OK
;
3584 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3586 BDRVRawState
*s
= bs
->opaque
;
3589 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3590 perror("CDROMEJECT");
3592 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3593 perror("CDROMEJECT");
3597 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3599 BDRVRawState
*s
= bs
->opaque
;
3601 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3603 * Note: an error can happen if the distribution automatically
3606 /* perror("CDROM_LOCKDOOR"); */
3610 static BlockDriver bdrv_host_cdrom
= {
3611 .format_name
= "host_cdrom",
3612 .protocol_name
= "host_cdrom",
3613 .instance_size
= sizeof(BDRVRawState
),
3614 .bdrv_needs_filename
= true,
3615 .bdrv_probe_device
= cdrom_probe_device
,
3616 .bdrv_parse_filename
= cdrom_parse_filename
,
3617 .bdrv_file_open
= cdrom_open
,
3618 .bdrv_close
= raw_close
,
3619 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3620 .bdrv_reopen_commit
= raw_reopen_commit
,
3621 .bdrv_reopen_abort
= raw_reopen_abort
,
3622 .bdrv_co_create_opts
= hdev_co_create_opts
,
3623 .create_opts
= &raw_create_opts
,
3624 .mutable_opts
= mutable_opts
,
3625 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3628 .bdrv_co_preadv
= raw_co_preadv
,
3629 .bdrv_co_pwritev
= raw_co_pwritev
,
3630 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3631 .bdrv_refresh_limits
= raw_refresh_limits
,
3632 .bdrv_io_plug
= raw_aio_plug
,
3633 .bdrv_io_unplug
= raw_aio_unplug
,
3634 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3636 .bdrv_co_truncate
= raw_co_truncate
,
3637 .bdrv_getlength
= raw_getlength
,
3638 .has_variable_length
= true,
3639 .bdrv_get_allocated_file_size
3640 = raw_get_allocated_file_size
,
3642 /* removable device support */
3643 .bdrv_is_inserted
= cdrom_is_inserted
,
3644 .bdrv_eject
= cdrom_eject
,
3645 .bdrv_lock_medium
= cdrom_lock_medium
,
3647 /* generic scsi device */
3648 .bdrv_co_ioctl
= hdev_co_ioctl
,
3650 #endif /* __linux__ */
3652 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3653 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3656 BDRVRawState
*s
= bs
->opaque
;
3657 Error
*local_err
= NULL
;
3662 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3664 error_propagate(errp
, local_err
);
3668 /* make sure the door isn't locked at this time */
3669 ioctl(s
->fd
, CDIOCALLOW
);
3673 static int cdrom_probe_device(const char *filename
)
3675 if (strstart(filename
, "/dev/cd", NULL
) ||
3676 strstart(filename
, "/dev/acd", NULL
))
3681 static int cdrom_reopen(BlockDriverState
*bs
)
3683 BDRVRawState
*s
= bs
->opaque
;
3687 * Force reread of possibly changed/newly loaded disc,
3688 * FreeBSD seems to not notice sometimes...
3692 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3699 /* make sure the door isn't locked at this time */
3700 ioctl(s
->fd
, CDIOCALLOW
);
3704 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3706 return raw_getlength(bs
) > 0;
3709 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3711 BDRVRawState
*s
= bs
->opaque
;
3716 (void) ioctl(s
->fd
, CDIOCALLOW
);
3719 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3720 perror("CDIOCEJECT");
3722 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3723 perror("CDIOCCLOSE");
3729 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3731 BDRVRawState
*s
= bs
->opaque
;
3735 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3737 * Note: an error can happen if the distribution automatically
3740 /* perror("CDROM_LOCKDOOR"); */
3744 static BlockDriver bdrv_host_cdrom
= {
3745 .format_name
= "host_cdrom",
3746 .protocol_name
= "host_cdrom",
3747 .instance_size
= sizeof(BDRVRawState
),
3748 .bdrv_needs_filename
= true,
3749 .bdrv_probe_device
= cdrom_probe_device
,
3750 .bdrv_parse_filename
= cdrom_parse_filename
,
3751 .bdrv_file_open
= cdrom_open
,
3752 .bdrv_close
= raw_close
,
3753 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3754 .bdrv_reopen_commit
= raw_reopen_commit
,
3755 .bdrv_reopen_abort
= raw_reopen_abort
,
3756 .bdrv_co_create_opts
= hdev_co_create_opts
,
3757 .create_opts
= &raw_create_opts
,
3758 .mutable_opts
= mutable_opts
,
3760 .bdrv_co_preadv
= raw_co_preadv
,
3761 .bdrv_co_pwritev
= raw_co_pwritev
,
3762 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3763 .bdrv_refresh_limits
= raw_refresh_limits
,
3764 .bdrv_io_plug
= raw_aio_plug
,
3765 .bdrv_io_unplug
= raw_aio_unplug
,
3766 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3768 .bdrv_co_truncate
= raw_co_truncate
,
3769 .bdrv_getlength
= raw_getlength
,
3770 .has_variable_length
= true,
3771 .bdrv_get_allocated_file_size
3772 = raw_get_allocated_file_size
,
3774 /* removable device support */
3775 .bdrv_is_inserted
= cdrom_is_inserted
,
3776 .bdrv_eject
= cdrom_eject
,
3777 .bdrv_lock_medium
= cdrom_lock_medium
,
3779 #endif /* __FreeBSD__ */
3781 static void bdrv_file_init(void)
3784 * Register all the drivers. Note that order is important, the driver
3785 * registered last will get probed first.
3787 bdrv_register(&bdrv_file
);
3788 bdrv_register(&bdrv_host_device
);
3790 bdrv_register(&bdrv_host_cdrom
);
3792 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3793 bdrv_register(&bdrv_host_cdrom
);
3797 block_init(bdrv_file_init
);