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 "qapi/error.h"
27
#include "qemu/cutils.h"
28
#include "qemu/error-report.h"
29
#include "block/block-io.h"
30
#include "block/block_int.h"
31
#include "qemu/module.h"
32
#include "qemu/option.h"
33
#include "qemu/units.h"
34
#include "qemu/memalign.h"
36
#include "block/thread-pool.h"
38
#include "block/raw-aio.h"
39
#include "qapi/qmp/qdict.h"
40
#include "qapi/qmp/qstring.h"
42
#include "scsi/pr-manager.h"
43
#include "scsi/constants.h"
45
#if defined(__APPLE__) && (__MACH__)
47
#if defined(HAVE_HOST_BLOCK_DEVICE)
51
#include <IOKit/IOKitLib.h>
52
#include <IOKit/IOBSD.h>
53
#include <IOKit/storage/IOMediaBSDClient.h>
54
#include <IOKit/storage/IOMedia.h>
55
#include <IOKit/storage/IOCDMedia.h>
56
//#include <IOKit/storage/IOCDTypes.h>
57
#include <IOKit/storage/IODVDMedia.h>
58
#include <CoreFoundation/CoreFoundation.h>
59
#endif /* defined(HAVE_HOST_BLOCK_DEVICE) */
63
#define _POSIX_PTHREAD_SEMANTICS 1
69
#include <sys/syscall.h>
71
#if defined(CONFIG_BLKZONED)
72
#include <linux/blkzoned.h>
74
#include <linux/cdrom.h>
77
#include <linux/hdreg.h>
78
#include <linux/magic.h>
84
#define FS_NOCOW_FL 0x00800000 /* Do not cow file */
87
#if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
88
#include <linux/falloc.h>
90
#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
97
#include <sys/disklabel.h>
102
#include <sys/ioctl.h>
103
#include <sys/disklabel.h>
109
#include <sys/ioctl.h>
110
#include <sys/diskslice.h>
113
/* OS X does not have O_DSYNC */
116
#define O_DSYNC O_SYNC
117
#elif defined(O_FSYNC)
118
#define O_DSYNC O_FSYNC
122
/* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
124
#define O_DIRECT O_DSYNC
130
#define MAX_BLOCKSIZE 4096
132
/* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
133
* leaving a few more bytes for its future use. */
134
#define RAW_LOCK_PERM_BASE 100
135
#define RAW_LOCK_SHARED_BASE 200
137
typedef struct BDRVRawState {
144
/* The current permissions. */
146
uint64_t shared_perm;
148
/* The perms bits whose corresponding bytes are already locked in
150
uint64_t locked_perm;
151
uint64_t locked_shared_perm;
153
uint64_t aio_max_batch;
156
int perm_change_flags;
157
BDRVReopenState *reopen_state;
160
bool has_write_zeroes:1;
161
bool use_linux_aio:1;
162
bool has_laio_fdsync:1;
163
bool use_linux_io_uring:1;
164
int page_cache_inconsistent; /* errno from fdatasync failure */
166
bool needs_alignment;
167
bool force_alignment;
169
bool check_cache_dropped;
171
uint64_t discard_nb_ok;
172
uint64_t discard_nb_failed;
173
uint64_t discard_bytes_ok;
179
typedef struct BDRVRawReopenState {
182
bool check_cache_dropped;
185
static int fd_open(BlockDriverState *bs)
187
BDRVRawState *s = bs->opaque;
189
/* this is just to ensure s->fd is sane (its called by io ops) */
196
static int64_t raw_getlength(BlockDriverState *bs);
198
typedef struct RawPosixAIOData {
199
BlockDriverState *bs;
220
PreallocMode prealloc;
224
unsigned int *nr_zones;
225
BlockZoneDescriptor *zones;
233
#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
234
static int cdrom_reopen(BlockDriverState *bs);
238
* Elide EAGAIN and EACCES details when failing to lock, as this
239
* indicates that the specified file region is already locked by
240
* another process, which is considered a common scenario.
242
#define raw_lock_error_setg_errno(errp, err, fmt, ...) \
244
if ((err) == EAGAIN || (err) == EACCES) { \
245
error_setg((errp), (fmt), ## __VA_ARGS__); \
247
error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \
251
#if defined(__NetBSD__)
252
static int raw_normalize_devicepath(const char **filename, Error **errp)
254
static char namebuf[PATH_MAX];
255
const char *dp, *fname;
259
dp = strrchr(fname, '/');
260
if (lstat(fname, &sb) < 0) {
261
error_setg_file_open(errp, errno, fname);
265
if (!S_ISBLK(sb.st_mode)) {
270
snprintf(namebuf, PATH_MAX, "r%s", fname);
272
snprintf(namebuf, PATH_MAX, "%.*s/r%s",
273
(int)(dp - fname), fname, dp + 1);
276
warn_report("%s is a block device, using %s", fname, *filename);
281
static int raw_normalize_devicepath(const char **filename, Error **errp)
288
* Get logical block size via ioctl. On success store it in @sector_size_p.
290
static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
292
unsigned int sector_size;
293
bool success = false;
297
static const unsigned long ioctl_list[] = {
301
#ifdef DKIOCGETBLOCKSIZE
304
#ifdef DIOCGSECTORSIZE
309
/* Try a few ioctls to get the right size */
310
for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
311
if (ioctl(fd, ioctl_list[i], §or_size) >= 0) {
312
*sector_size_p = sector_size;
317
return success ? 0 : -errno;
321
* Get physical block size of @fd.
322
* On success, store it in @blk_size and return 0.
323
* On failure, return -errno.
325
static int probe_physical_blocksize(int fd, unsigned int *blk_size)
328
if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
338
* Returns true if no alignment restrictions are necessary even for files
339
* opened with O_DIRECT.
341
* raw_probe_alignment() probes the required alignment and assume that 1 means
342
* the probing failed, so it falls back to a safe default of 4k. This can be
343
* avoided if we know that byte alignment is okay for the file.
345
static bool dio_byte_aligned(int fd)
351
ret = fstatfs(fd, &buf);
352
if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) {
359
static bool raw_needs_alignment(BlockDriverState *bs)
361
BDRVRawState *s = bs->opaque;
363
if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) {
367
return s->force_alignment;
370
/* Check if read is allowed with given memory buffer and length.
372
* This function is used to check O_DIRECT memory buffer and request alignment.
374
static bool raw_is_io_aligned(int fd, void *buf, size_t len)
376
ssize_t ret = pread(fd, buf, len, 0);
383
/* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
384
* other errors (e.g. real I/O error), which could happen on a failed
385
* drive, since we only care about probing alignment.
387
if (errno != EINVAL) {
395
static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
397
BDRVRawState *s = bs->opaque;
399
size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size());
400
size_t alignments[] = {1, 512, 1024, 2048, 4096};
402
/* For SCSI generic devices the alignment is not really used.
403
With buffered I/O, we don't have any restrictions. */
404
if (bdrv_is_sg(bs) || !s->needs_alignment) {
405
bs->bl.request_alignment = 1;
410
bs->bl.request_alignment = 0;
412
/* Let's try to use the logical blocksize for the alignment. */
413
if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
414
bs->bl.request_alignment = 0;
419
* The XFS ioctl definitions are shipped in extra packages that might
420
* not always be available. Since we just need the XFS_IOC_DIOINFO ioctl
421
* here, we simply use our own definition instead:
428
if (ioctl(fd, _IOR('X', 30, struct xfs_dioattr), &da) >= 0) {
429
bs->bl.request_alignment = da.d_miniosz;
430
/* The kernel returns wrong information for d_mem */
431
/* s->buf_align = da.d_mem; */
436
* If we could not get the sizes so far, we can only guess them. First try
437
* to detect request alignment, since it is more likely to succeed. Then
438
* try to detect buf_align, which cannot be detected in some cases (e.g.
439
* Gluster). If buf_align cannot be detected, we fallback to the value of
443
if (!bs->bl.request_alignment) {
446
buf = qemu_memalign(max_align, max_align);
447
for (i = 0; i < ARRAY_SIZE(alignments); i++) {
448
align = alignments[i];
449
if (raw_is_io_aligned(fd, buf, align)) {
450
/* Fallback to safe value. */
451
bs->bl.request_alignment = (align != 1) ? align : max_align;
461
buf = qemu_memalign(max_align, 2 * max_align);
462
for (i = 0; i < ARRAY_SIZE(alignments); i++) {
463
align = alignments[i];
464
if (raw_is_io_aligned(fd, buf + align, max_align)) {
465
/* Fallback to request_alignment. */
466
s->buf_align = (align != 1) ? align : bs->bl.request_alignment;
473
if (!s->buf_align || !bs->bl.request_alignment) {
474
error_setg(errp, "Could not find working O_DIRECT alignment");
475
error_append_hint(errp, "Try cache.direct=off\n");
479
static int check_hdev_writable(int fd)
482
/* Linux block devices can be configured "read-only" using blockdev(8).
483
* This is independent of device node permissions and therefore open(2)
484
* with O_RDWR succeeds. Actual writes fail with EPERM.
486
* bdrv_open() is supposed to fail if the disk is read-only. Explicitly
487
* check for read-only block devices so that Linux block devices behave
493
if (fstat(fd, &st)) {
497
if (!S_ISBLK(st.st_mode)) {
501
if (ioctl(fd, BLKROGET, &readonly) < 0) {
508
#endif /* defined(BLKROGET) */
512
static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers)
514
bool read_write = false;
515
assert(open_flags != NULL);
517
*open_flags |= O_BINARY;
518
*open_flags &= ~O_ACCMODE;
520
if (bdrv_flags & BDRV_O_AUTO_RDONLY) {
521
read_write = has_writers;
522
} else if (bdrv_flags & BDRV_O_RDWR) {
527
*open_flags |= O_RDWR;
529
*open_flags |= O_RDONLY;
532
/* Use O_DSYNC for write-through caching, no flags for write-back caching,
533
* and O_DIRECT for no caching. */
534
if ((bdrv_flags & BDRV_O_NOCACHE)) {
535
*open_flags |= O_DIRECT;
539
static void raw_parse_filename(const char *filename, QDict *options,
542
bdrv_parse_filename_strip_prefix(filename, "file:", options);
545
static QemuOptsList raw_runtime_opts = {
547
.head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
551
.type = QEMU_OPT_STRING,
552
.help = "File name of the image",
556
.type = QEMU_OPT_STRING,
557
.help = "host AIO implementation (threads, native, io_uring)",
560
.name = "aio-max-batch",
561
.type = QEMU_OPT_NUMBER,
562
.help = "AIO max batch size (0 = auto handled by AIO backend, default: 0)",
566
.type = QEMU_OPT_STRING,
567
.help = "file locking mode (on/off/auto, default: auto)",
570
.name = "pr-manager",
571
.type = QEMU_OPT_STRING,
572
.help = "id of persistent reservation manager object (default: none)",
574
#if defined(__linux__)
576
.name = "drop-cache",
577
.type = QEMU_OPT_BOOL,
578
.help = "invalidate page cache during live migration (default: on)",
582
.name = "x-check-cache-dropped",
583
.type = QEMU_OPT_BOOL,
584
.help = "check that page cache was dropped on live migration (default: off)"
586
{ /* end of list */ }
590
static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL };
592
static int raw_open_common(BlockDriverState *bs, QDict *options,
593
int bdrv_flags, int open_flags,
594
bool device, Error **errp)
596
BDRVRawState *s = bs->opaque;
598
Error *local_err = NULL;
599
const char *filename = NULL;
601
BlockdevAioOptions aio, aio_default;
606
opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
607
if (!qemu_opts_absorb_qdict(opts, options, errp)) {
612
filename = qemu_opt_get(opts, "filename");
614
ret = raw_normalize_devicepath(&filename, errp);
619
if (bdrv_flags & BDRV_O_NATIVE_AIO) {
620
aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE;
621
#ifdef CONFIG_LINUX_IO_URING
622
} else if (bdrv_flags & BDRV_O_IO_URING) {
623
aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING;
626
aio_default = BLOCKDEV_AIO_OPTIONS_THREADS;
629
aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
630
qemu_opt_get(opts, "aio"),
631
aio_default, &local_err);
633
error_propagate(errp, local_err);
638
s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
639
#ifdef CONFIG_LINUX_IO_URING
640
s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING);
643
s->aio_max_batch = qemu_opt_get_number(opts, "aio-max-batch", 0);
645
locking = qapi_enum_parse(&OnOffAuto_lookup,
646
qemu_opt_get(opts, "locking"),
647
ON_OFF_AUTO_AUTO, &local_err);
649
error_propagate(errp, local_err);
656
if (!qemu_has_ofd_lock()) {
657
warn_report("File lock requested but OFD locking syscall is "
658
"unavailable, falling back to POSIX file locks");
659
error_printf("Due to the implementation, locks can be lost "
663
case ON_OFF_AUTO_OFF:
666
case ON_OFF_AUTO_AUTO:
667
s->use_lock = qemu_has_ofd_lock();
673
str = qemu_opt_get(opts, "pr-manager");
675
s->pr_mgr = pr_manager_lookup(str, &local_err);
677
error_propagate(errp, local_err);
683
s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true);
684
s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped",
687
s->open_flags = open_flags;
688
raw_parse_flags(bdrv_flags, &s->open_flags, false);
691
fd = qemu_open(filename, s->open_flags, errp);
692
ret = fd < 0 ? -errno : 0;
702
/* Check s->open_flags rather than bdrv_flags due to auto-read-only */
703
if (s->open_flags & O_RDWR) {
704
ret = check_hdev_writable(s->fd);
706
error_setg_errno(errp, -ret, "The device is not writable");
712
s->shared_perm = BLK_PERM_ALL;
714
#ifdef CONFIG_LINUX_AIO
715
/* Currently Linux does AIO only for files opened with O_DIRECT */
716
if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) {
717
error_setg(errp, "aio=native was specified, but it requires "
718
"cache.direct=on, which was not specified.");
722
if (s->use_linux_aio) {
723
s->has_laio_fdsync = laio_has_fdsync(s->fd);
726
if (s->use_linux_aio) {
727
error_setg(errp, "aio=native was specified, but is not supported "
732
#endif /* !defined(CONFIG_LINUX_AIO) */
734
#ifndef CONFIG_LINUX_IO_URING
735
if (s->use_linux_io_uring) {
736
error_setg(errp, "aio=io_uring was specified, but is not supported "
741
#endif /* !defined(CONFIG_LINUX_IO_URING) */
743
s->has_discard = true;
744
s->has_write_zeroes = true;
746
if (fstat(s->fd, &st) < 0) {
748
error_setg_errno(errp, errno, "Could not stat file");
753
if (!S_ISREG(st.st_mode)) {
754
error_setg(errp, "'%s' driver requires '%s' to be a regular file",
755
bs->drv->format_name, bs->filename);
759
s->has_fallocate = true;
762
if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
763
error_setg(errp, "'%s' driver requires '%s' to be either "
764
"a character or block device",
765
bs->drv->format_name, bs->filename);
770
#ifdef CONFIG_BLKZONED
772
* The kernel page cache does not reliably work for writes to SWR zones
773
* of zoned block device because it can not guarantee the order of writes.
775
if ((bs->bl.zoned != BLK_Z_NONE) &&
776
(!(s->open_flags & O_DIRECT))) {
777
error_setg(errp, "The driver supports zoned devices, and it requires "
778
"cache.direct=on, which was not specified.");
779
return -EINVAL; /* No host kernel page cache */
783
if (S_ISBLK(st.st_mode)) {
785
/* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
786
* not rely on the contents of discarded blocks unless using O_DIRECT.
787
* Same for BLKZEROOUT.
789
if (!(bs->open_flags & BDRV_O_NOCACHE)) {
790
s->has_write_zeroes = false;
795
if (S_ISCHR(st.st_mode)) {
797
* The file is a char device (disk), which on FreeBSD isn't behind
798
* a pager, so force all requests to be aligned. This is needed
799
* so QEMU makes sure all IO operations on the device are aligned
800
* to sector size, or else FreeBSD will reject them with EINVAL.
802
s->force_alignment = true;
805
s->needs_alignment = raw_needs_alignment(bs);
807
bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK;
808
if (S_ISREG(st.st_mode)) {
809
/* When extending regular files, we get zeros from the OS */
810
bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
814
if (ret < 0 && s->fd != -1) {
817
if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
824
static int raw_open(BlockDriverState *bs, QDict *options, int flags,
827
BDRVRawState *s = bs->opaque;
829
s->type = FTYPE_FILE;
830
return raw_open_common(bs, options, flags, 0, false, errp);
839
#define PERM_FOREACH(i) \
840
for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
842
/* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
843
* file; if @unlock == true, also unlock the unneeded bytes.
844
* @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
846
static int raw_apply_lock_bytes(BDRVRawState *s, int fd,
847
uint64_t perm_lock_bits,
848
uint64_t shared_perm_lock_bits,
849
bool unlock, Error **errp)
853
uint64_t locked_perm, locked_shared_perm;
856
locked_perm = s->locked_perm;
857
locked_shared_perm = s->locked_shared_perm;
860
* We don't have the previous bits, just lock/unlock for each of the
864
locked_perm = BLK_PERM_ALL;
865
locked_shared_perm = BLK_PERM_ALL;
868
locked_shared_perm = 0;
873
int off = RAW_LOCK_PERM_BASE + i;
874
uint64_t bit = (1ULL << i);
875
if ((perm_lock_bits & bit) && !(locked_perm & bit)) {
876
ret = qemu_lock_fd(fd, off, 1, false);
878
raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d",
882
s->locked_perm |= bit;
884
} else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) {
885
ret = qemu_unlock_fd(fd, off, 1);
887
error_setg_errno(errp, -ret, "Failed to unlock byte %d", off);
890
s->locked_perm &= ~bit;
895
int off = RAW_LOCK_SHARED_BASE + i;
896
uint64_t bit = (1ULL << i);
897
if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) {
898
ret = qemu_lock_fd(fd, off, 1, false);
900
raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d",
904
s->locked_shared_perm |= bit;
906
} else if (unlock && (locked_shared_perm & bit) &&
907
!(shared_perm_lock_bits & bit)) {
908
ret = qemu_unlock_fd(fd, off, 1);
910
error_setg_errno(errp, -ret, "Failed to unlock byte %d", off);
913
s->locked_shared_perm &= ~bit;
920
/* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
921
static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm,
928
int off = RAW_LOCK_SHARED_BASE + i;
929
uint64_t p = 1ULL << i;
931
ret = qemu_lock_fd_test(fd, off, 1, true);
933
char *perm_name = bdrv_perm_names(p);
935
raw_lock_error_setg_errno(errp, -ret,
936
"Failed to get \"%s\" lock",
944
int off = RAW_LOCK_PERM_BASE + i;
945
uint64_t p = 1ULL << i;
946
if (!(shared_perm & p)) {
947
ret = qemu_lock_fd_test(fd, off, 1, true);
949
char *perm_name = bdrv_perm_names(p);
951
raw_lock_error_setg_errno(errp, -ret,
952
"Failed to get shared \"%s\" lock",
962
static int raw_handle_perm_lock(BlockDriverState *bs,
964
uint64_t new_perm, uint64_t new_shared,
967
BDRVRawState *s = bs->opaque;
969
Error *local_err = NULL;
975
if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
981
if ((s->perm | new_perm) == s->perm &&
982
(s->shared_perm & new_shared) == s->shared_perm)
985
* We are going to unlock bytes, it should not fail. If it fail due
986
* to some fs-dependent permission-unrelated reasons (which occurs
987
* sometimes on NFS and leads to abort in bdrv_replace_child) we
988
* can't prevent such errors by any check here. And we ignore them
989
* anyway in ABORT and COMMIT.
993
ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm,
994
~s->shared_perm | ~new_shared,
997
ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp);
1001
error_append_hint(errp,
1002
"Is another process using the image [%s]?\n",
1005
/* fall through to unlock bytes. */
1007
raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm,
1010
/* Theoretically the above call only unlocks bytes and it cannot
1011
* fail. Something weird happened, report it.
1013
warn_report_err(local_err);
1017
raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared,
1020
/* Theoretically the above call only unlocks bytes and it cannot
1021
* fail. Something weird happened, report it.
1023
warn_report_err(local_err);
1030
/* Sets a specific flag */
1031
static int fcntl_setfl(int fd, int flag)
1035
flags = fcntl(fd, F_GETFL);
1039
if (fcntl(fd, F_SETFL, flags | flag) == -1) {
1045
static int raw_reconfigure_getfd(BlockDriverState *bs, int flags,
1046
int *open_flags, uint64_t perm, Error **errp)
1048
BDRVRawState *s = bs->opaque;
1051
bool has_writers = perm &
1052
(BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE);
1053
int fcntl_flags = O_APPEND | O_NONBLOCK;
1055
fcntl_flags |= O_NOATIME;
1059
if (s->type == FTYPE_CD) {
1060
*open_flags |= O_NONBLOCK;
1063
raw_parse_flags(flags, open_flags, has_writers);
1066
/* Not all operating systems have O_ASYNC, and those that don't
1067
* will not let us track the state into rs->open_flags (typically
1068
* you achieve the same effect with an ioctl, for example I_SETSIG
1069
* on Solaris). But we do not use O_ASYNC, so that's fine.
1071
assert((s->open_flags & O_ASYNC) == 0);
1074
if (*open_flags == s->open_flags) {
1075
/* We're lucky, the existing fd is fine */
1079
if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
1080
/* dup the original fd */
1081
fd = qemu_dup(s->fd);
1083
ret = fcntl_setfl(fd, *open_flags);
1091
/* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
1093
const char *normalized_filename = bs->filename;
1094
ret = raw_normalize_devicepath(&normalized_filename, errp);
1096
fd = qemu_open(normalized_filename, *open_flags, errp);
1103
if (fd != -1 && (*open_flags & O_RDWR)) {
1104
ret = check_hdev_writable(fd);
1107
error_setg_errno(errp, -ret, "The device is not writable");
1115
static int raw_reopen_prepare(BDRVReopenState *state,
1116
BlockReopenQueue *queue, Error **errp)
1119
BDRVRawReopenState *rs;
1123
assert(state != NULL);
1124
assert(state->bs != NULL);
1126
s = state->bs->opaque;
1128
state->opaque = g_new0(BDRVRawReopenState, 1);
1131
/* Handle options changes */
1132
opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
1133
if (!qemu_opts_absorb_qdict(opts, state->options, errp)) {
1138
rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true);
1139
rs->check_cache_dropped =
1140
qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false);
1142
/* This driver's reopen function doesn't currently allow changing
1143
* other options, so let's put them back in the original QDict and
1144
* bdrv_reopen_prepare() will detect changes and complain. */
1145
qemu_opts_to_qdict(opts, state->options);
1148
* As part of reopen prepare we also want to create new fd by
1149
* raw_reconfigure_getfd(). But it wants updated "perm", when in
1150
* bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to
1151
* permission update. Happily, permission update is always a part
1152
* (a separate stage) of bdrv_reopen_multiple() so we can rely on this
1153
* fact and reconfigure fd in raw_check_perm().
1156
s->reopen_state = state;
1160
qemu_opts_del(opts);
1164
static void raw_reopen_commit(BDRVReopenState *state)
1166
BDRVRawReopenState *rs = state->opaque;
1167
BDRVRawState *s = state->bs->opaque;
1169
s->drop_cache = rs->drop_cache;
1170
s->check_cache_dropped = rs->check_cache_dropped;
1171
s->open_flags = rs->open_flags;
1172
g_free(state->opaque);
1173
state->opaque = NULL;
1175
assert(s->reopen_state == state);
1176
s->reopen_state = NULL;
1180
static void raw_reopen_abort(BDRVReopenState *state)
1182
BDRVRawReopenState *rs = state->opaque;
1183
BDRVRawState *s = state->bs->opaque;
1185
/* nothing to do if NULL, we didn't get far enough */
1190
g_free(state->opaque);
1191
state->opaque = NULL;
1193
assert(s->reopen_state == state);
1194
s->reopen_state = NULL;
1197
static int hdev_get_max_hw_transfer(int fd, struct stat *st)
1200
if (S_ISBLK(st->st_mode)) {
1201
unsigned short max_sectors = 0;
1202
if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
1203
return max_sectors * 512;
1207
if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
1218
* Get a sysfs attribute value as character string.
1221
static int get_sysfs_str_val(struct stat *st, const char *attribute,
1223
g_autofree char *sysfspath = NULL;
1226
if (!S_ISBLK(st->st_mode)) {
1230
sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/%s",
1231
major(st->st_rdev), minor(st->st_rdev),
1233
if (!g_file_get_contents(sysfspath, val, &len, NULL)) {
1237
/* The file is ended with '\n' */
1240
if (*(p + len - 1) == '\n') {
1241
*(p + len - 1) = '\0';
1247
#if defined(CONFIG_BLKZONED)
1248
static int get_sysfs_zoned_model(struct stat *st, BlockZoneModel *zoned)
1250
g_autofree char *val = NULL;
1253
ret = get_sysfs_str_val(st, "zoned", &val);
1258
if (strcmp(val, "host-managed") == 0) {
1260
} else if (strcmp(val, "host-aware") == 0) {
1262
} else if (strcmp(val, "none") == 0) {
1263
*zoned = BLK_Z_NONE;
1269
#endif /* defined(CONFIG_BLKZONED) */
1272
* Get a sysfs attribute value as a long integer.
1275
static long get_sysfs_long_val(struct stat *st, const char *attribute)
1277
g_autofree char *str = NULL;
1282
ret = get_sysfs_str_val(st, attribute, &str);
1287
/* The file is ended with '\n', pass 'end' to accept that. */
1288
ret = qemu_strtol(str, &end, 10, &val);
1289
if (ret == 0 && end && *end == '\0') {
1296
static int hdev_get_max_segments(int fd, struct stat *st)
1301
if (S_ISCHR(st->st_mode)) {
1302
if (ioctl(fd, SG_GET_SG_TABLESIZE, &ret) == 0) {
1307
return get_sysfs_long_val(st, "max_segments");
1313
#if defined(CONFIG_BLKZONED)
1315
* If the reset_all flag is true, then the wps of zone whose state is
1316
* not readonly or offline should be all reset to the start sector.
1317
* Else, take the real wp of the device.
1319
static int get_zones_wp(BlockDriverState *bs, int fd, int64_t offset,
1320
unsigned int nrz, bool reset_all)
1322
struct blk_zone *blkz;
1324
uint64_t sector = offset >> BDRV_SECTOR_BITS;
1325
BlockZoneWps *wps = bs->wps;
1326
unsigned int j = offset / bs->bl.zone_size;
1327
unsigned int n = 0, i = 0;
1329
rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct blk_zone);
1330
g_autofree struct blk_zone_report *rep = NULL;
1332
rep = g_malloc(rep_size);
1333
blkz = (struct blk_zone *)(rep + 1);
1335
memset(rep, 0, rep_size);
1336
rep->sector = sector;
1337
rep->nr_zones = nrz - n;
1340
ret = ioctl(fd, BLKREPORTZONE, rep);
1341
} while (ret != 0 && errno == EINTR);
1343
error_report("%d: ioctl BLKREPORTZONE at %" PRId64 " failed %d",
1348
if (!rep->nr_zones) {
1352
for (i = 0; i < rep->nr_zones; ++i, ++n, ++j) {
1354
* The wp tracking cares only about sequential writes required and
1355
* sequential write preferred zones so that the wp can advance to
1356
* the right location.
1357
* Use the most significant bit of the wp location to indicate the
1358
* zone type: 0 for SWR/SWP zones and 1 for conventional zones.
1360
if (blkz[i].type == BLK_ZONE_TYPE_CONVENTIONAL) {
1361
wps->wp[j] |= 1ULL << 63;
1363
switch(blkz[i].cond) {
1364
case BLK_ZONE_COND_FULL:
1365
case BLK_ZONE_COND_READONLY:
1366
/* Zone not writable */
1367
wps->wp[j] = (blkz[i].start + blkz[i].len) << BDRV_SECTOR_BITS;
1369
case BLK_ZONE_COND_OFFLINE:
1370
/* Zone not writable nor readable */
1371
wps->wp[j] = (blkz[i].start) << BDRV_SECTOR_BITS;
1375
wps->wp[j] = blkz[i].start << BDRV_SECTOR_BITS;
1377
wps->wp[j] = blkz[i].wp << BDRV_SECTOR_BITS;
1383
sector = blkz[i - 1].start + blkz[i - 1].len;
1389
static void update_zones_wp(BlockDriverState *bs, int fd, int64_t offset,
1392
if (get_zones_wp(bs, fd, offset, nrz, 0) < 0) {
1393
error_report("update zone wp failed");
1397
static void raw_refresh_zoned_limits(BlockDriverState *bs, struct stat *st,
1400
BDRVRawState *s = bs->opaque;
1401
BlockZoneModel zoned;
1404
ret = get_sysfs_zoned_model(st, &zoned);
1405
if (ret < 0 || zoned == BLK_Z_NONE) {
1408
bs->bl.zoned = zoned;
1410
ret = get_sysfs_long_val(st, "max_open_zones");
1412
bs->bl.max_open_zones = ret;
1415
ret = get_sysfs_long_val(st, "max_active_zones");
1417
bs->bl.max_active_zones = ret;
1421
* The zoned device must at least have zone size and nr_zones fields.
1423
ret = get_sysfs_long_val(st, "chunk_sectors");
1425
error_setg_errno(errp, -ret, "Unable to read chunk_sectors "
1429
error_setg(errp, "Read 0 from chunk_sectors sysfs attribute");
1432
bs->bl.zone_size = ret << BDRV_SECTOR_BITS;
1434
ret = get_sysfs_long_val(st, "nr_zones");
1436
error_setg_errno(errp, -ret, "Unable to read nr_zones "
1440
error_setg(errp, "Read 0 from nr_zones sysfs attribute");
1443
bs->bl.nr_zones = ret;
1445
ret = get_sysfs_long_val(st, "zone_append_max_bytes");
1447
bs->bl.max_append_sectors = ret >> BDRV_SECTOR_BITS;
1450
ret = get_sysfs_long_val(st, "physical_block_size");
1452
bs->bl.write_granularity = ret;
1455
/* The refresh_limits() function can be called multiple times. */
1457
bs->wps = g_malloc(sizeof(BlockZoneWps) +
1458
sizeof(int64_t) * bs->bl.nr_zones);
1459
ret = get_zones_wp(bs, s->fd, 0, bs->bl.nr_zones, 0);
1461
error_setg_errno(errp, -ret, "report wps failed");
1464
qemu_co_mutex_init(&bs->wps->colock);
1468
bs->bl.zoned = BLK_Z_NONE;
1472
#else /* !defined(CONFIG_BLKZONED) */
1473
static void raw_refresh_zoned_limits(BlockDriverState *bs, struct stat *st,
1476
bs->bl.zoned = BLK_Z_NONE;
1478
#endif /* !defined(CONFIG_BLKZONED) */
1480
static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
1482
BDRVRawState *s = bs->opaque;
1485
s->needs_alignment = raw_needs_alignment(bs);
1486
raw_probe_alignment(bs, s->fd, errp);
1488
bs->bl.min_mem_alignment = s->buf_align;
1489
bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size());
1492
* Maximum transfers are best effort, so it is okay to ignore any
1493
* errors. That said, based on the man page errors in fstat would be
1494
* very much unexpected; the only possible case seems to be ENOMEM.
1496
if (fstat(s->fd, &st)) {
1500
#if defined(__APPLE__) && (__MACH__)
1503
if (!fstatfs(s->fd, &buf)) {
1504
bs->bl.opt_transfer = buf.f_iosize;
1505
bs->bl.pdiscard_alignment = buf.f_bsize;
1509
if (bdrv_is_sg(bs) || S_ISBLK(st.st_mode)) {
1510
int ret = hdev_get_max_hw_transfer(s->fd, &st);
1512
if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
1513
bs->bl.max_hw_transfer = ret;
1516
ret = hdev_get_max_segments(s->fd, &st);
1518
bs->bl.max_hw_iov = ret;
1522
raw_refresh_zoned_limits(bs, &st, errp);
1525
static int check_for_dasd(int fd)
1528
struct dasd_information2_t info = {0};
1530
return ioctl(fd, BIODASDINFO2, &info);
1537
* Try to get @bs's logical and physical block size.
1538
* On success, store them in @bsz and return zero.
1539
* On failure, return negative errno.
1541
static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
1543
BDRVRawState *s = bs->opaque;
1546
/* If DASD or zoned devices, get blocksizes */
1547
if (check_for_dasd(s->fd) < 0) {
1548
/* zoned devices are not DASD */
1549
if (bs->bl.zoned == BLK_Z_NONE) {
1553
ret = probe_logical_blocksize(s->fd, &bsz->log);
1557
return probe_physical_blocksize(s->fd, &bsz->phys);
1561
* Try to get @bs's geometry: cyls, heads, sectors.
1562
* On success, store them in @geo and return 0.
1563
* On failure return -errno.
1564
* (Allows block driver to assign default geometry values that guest sees)
1567
static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1569
BDRVRawState *s = bs->opaque;
1570
struct hd_geometry ioctl_geo = {0};
1572
/* If DASD, get its geometry */
1573
if (check_for_dasd(s->fd) < 0) {
1576
if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1579
/* HDIO_GETGEO may return success even though geo contains zeros
1580
(e.g. certain multipath setups) */
1581
if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1584
/* Do not return a geometry for partition */
1585
if (ioctl_geo.start != 0) {
1588
geo->heads = ioctl_geo.heads;
1589
geo->sectors = ioctl_geo.sectors;
1590
geo->cylinders = ioctl_geo.cylinders;
1594
#else /* __linux__ */
1595
static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1601
#if defined(__linux__)
1602
static int handle_aiocb_ioctl(void *opaque)
1604
RawPosixAIOData *aiocb = opaque;
1607
ret = RETRY_ON_EINTR(
1608
ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf)
1618
static int handle_aiocb_flush(void *opaque)
1620
RawPosixAIOData *aiocb = opaque;
1621
BDRVRawState *s = aiocb->bs->opaque;
1624
if (s->page_cache_inconsistent) {
1625
return -s->page_cache_inconsistent;
1628
ret = qemu_fdatasync(aiocb->aio_fildes);
1630
trace_file_flush_fdatasync_failed(errno);
1632
/* There is no clear definition of the semantics of a failing fsync(),
1633
* so we may have to assume the worst. The sad truth is that this
1634
* assumption is correct for Linux. Some pages are now probably marked
1635
* clean in the page cache even though they are inconsistent with the
1636
* on-disk contents. The next fdatasync() call would succeed, but no
1637
* further writeback attempt will be made. We can't get back to a state
1638
* in which we know what is on disk (we would have to rewrite
1639
* everything that was touched since the last fdatasync() at least), so
1640
* make bdrv_flush() fail permanently. Given that the behaviour isn't
1641
* really defined, I have little hope that other OSes are doing better.
1643
* Obviously, this doesn't affect O_DIRECT, which bypasses the page
1645
if ((s->open_flags & O_DIRECT) == 0) {
1646
s->page_cache_inconsistent = errno;
1655
static bool preadv_present = true;
1658
qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1660
return preadv(fd, iov, nr_iov, offset);
1664
qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1666
return pwritev(fd, iov, nr_iov, offset);
1671
static bool preadv_present = false;
1674
qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1680
qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1687
static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1691
len = RETRY_ON_EINTR(
1692
(aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) ?
1693
qemu_pwritev(aiocb->aio_fildes,
1696
aiocb->aio_offset) :
1697
qemu_preadv(aiocb->aio_fildes,
1710
* Read/writes the data to/from a given linear buffer.
1712
* Returns the number of bytes handles or -errno in case of an error. Short
1713
* reads are only returned if the end of the file is reached.
1715
static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1720
while (offset < aiocb->aio_nbytes) {
1721
if (aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) {
1722
len = pwrite(aiocb->aio_fildes,
1723
(const char *)buf + offset,
1724
aiocb->aio_nbytes - offset,
1725
aiocb->aio_offset + offset);
1727
len = pread(aiocb->aio_fildes,
1729
aiocb->aio_nbytes - offset,
1730
aiocb->aio_offset + offset);
1732
if (len == -1 && errno == EINTR) {
1734
} else if (len == -1 && errno == EINVAL &&
1735
(aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1736
!(aiocb->aio_type & QEMU_AIO_WRITE) &&
1738
/* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1739
* after a short read. Assume that O_DIRECT short reads only occur
1740
* at EOF. Therefore this is a short read, not an I/O error.
1743
} else if (len == -1) {
1746
} else if (len == 0) {
1755
static int handle_aiocb_rw(void *opaque)
1757
RawPosixAIOData *aiocb = opaque;
1761
if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1763
* If there is just a single buffer, and it is properly aligned
1764
* we can just use plain pread/pwrite without any problems.
1766
if (aiocb->io.niov == 1) {
1767
nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base);
1771
* We have more than one iovec, and all are properly aligned.
1773
* Try preadv/pwritev first and fall back to linearizing the
1774
* buffer if it's not supported.
1776
if (preadv_present) {
1777
nbytes = handle_aiocb_rw_vector(aiocb);
1778
if (nbytes == aiocb->aio_nbytes ||
1779
(nbytes < 0 && nbytes != -ENOSYS)) {
1782
preadv_present = false;
1786
* XXX(hch): short read/write. no easy way to handle the reminder
1787
* using these interfaces. For now retry using plain
1793
* Ok, we have to do it the hard way, copy all segments into
1794
* a single aligned buffer.
1796
buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1802
if (aiocb->aio_type & QEMU_AIO_WRITE) {
1806
for (i = 0; i < aiocb->io.niov; ++i) {
1807
memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len);
1808
p += aiocb->io.iov[i].iov_len;
1810
assert(p - buf == aiocb->aio_nbytes);
1813
nbytes = handle_aiocb_rw_linear(aiocb, buf);
1814
if (!(aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND))) {
1816
size_t count = aiocb->aio_nbytes, copy;
1819
for (i = 0; i < aiocb->io.niov && count; ++i) {
1821
if (copy > aiocb->io.iov[i].iov_len) {
1822
copy = aiocb->io.iov[i].iov_len;
1824
memcpy(aiocb->io.iov[i].iov_base, p, copy);
1825
assert(count >= copy);
1834
if (nbytes == aiocb->aio_nbytes) {
1836
} else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) {
1837
if (aiocb->aio_type & QEMU_AIO_WRITE) {
1840
iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes,
1841
0, aiocb->aio_nbytes - nbytes);
1850
#if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD)
1851
static int translate_err(int err)
1853
if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1861
#ifdef CONFIG_FALLOCATE
1862
static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1865
if (fallocate(fd, mode, offset, len) == 0) {
1868
} while (errno == EINTR);
1869
return translate_err(-errno);
1873
static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1876
BDRVRawState *s = aiocb->bs->opaque;
1878
if (!s->has_write_zeroes) {
1883
/* The BLKZEROOUT implementation in the kernel doesn't set
1884
* BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1886
if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) {
1888
uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1889
if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1892
} while (errno == EINTR);
1894
ret = translate_err(-errno);
1895
if (ret == -ENOTSUP) {
1896
s->has_write_zeroes = false;
1904
static int handle_aiocb_write_zeroes(void *opaque)
1906
RawPosixAIOData *aiocb = opaque;
1907
#ifdef CONFIG_FALLOCATE
1908
BDRVRawState *s = aiocb->bs->opaque;
1912
if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1913
return handle_aiocb_write_zeroes_block(aiocb);
1916
#ifdef CONFIG_FALLOCATE_ZERO_RANGE
1917
if (s->has_write_zeroes) {
1918
int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1919
aiocb->aio_offset, aiocb->aio_nbytes);
1920
if (ret == -ENOTSUP) {
1921
s->has_write_zeroes = false;
1922
} else if (ret == 0 || ret != -EINVAL) {
1926
* Note: Some file systems do not like unaligned byte ranges, and
1927
* return EINVAL in such a case, though they should not do it according
1928
* to the man-page of fallocate(). Thus we simply ignore this return
1929
* value and try the other fallbacks instead.
1934
#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1935
if (s->has_discard && s->has_fallocate) {
1936
int ret = do_fallocate(s->fd,
1937
FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1938
aiocb->aio_offset, aiocb->aio_nbytes);
1940
ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1941
if (ret == 0 || ret != -ENOTSUP) {
1944
s->has_fallocate = false;
1945
} else if (ret == -EINVAL) {
1947
* Some file systems like older versions of GPFS do not like un-
1948
* aligned byte ranges, and return EINVAL in such a case, though
1949
* they should not do it according to the man-page of fallocate().
1950
* Warn about the bad filesystem and try the final fallback instead.
1952
warn_report_once("Your file system is misbehaving: "
1953
"fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. "
1954
"Please report this bug to your file system "
1956
} else if (ret != -ENOTSUP) {
1959
s->has_discard = false;
1964
#ifdef CONFIG_FALLOCATE
1965
/* Last resort: we are trying to extend the file with zeroed data. This
1966
* can be done via fallocate(fd, 0) */
1967
len = raw_getlength(aiocb->bs);
1968
if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
1969
int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1970
if (ret == 0 || ret != -ENOTSUP) {
1973
s->has_fallocate = false;
1980
static int handle_aiocb_write_zeroes_unmap(void *opaque)
1982
RawPosixAIOData *aiocb = opaque;
1983
BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque;
1985
/* First try to write zeros and unmap at the same time */
1987
#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1988
int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1989
aiocb->aio_offset, aiocb->aio_nbytes);
2000
/* If we couldn't manage to unmap while guaranteed that the area reads as
2001
* all-zero afterwards, just write zeroes without unmapping */
2002
return handle_aiocb_write_zeroes(aiocb);
2005
#ifndef HAVE_COPY_FILE_RANGE
2006
static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd,
2007
off_t *out_off, size_t len, unsigned int flags)
2009
#ifdef __NR_copy_file_range
2010
return syscall(__NR_copy_file_range, in_fd, in_off, out_fd,
2011
out_off, len, flags);
2020
* parse_zone - Fill a zone descriptor
2022
#if defined(CONFIG_BLKZONED)
2023
static inline int parse_zone(struct BlockZoneDescriptor *zone,
2024
const struct blk_zone *blkz) {
2025
zone->start = blkz->start << BDRV_SECTOR_BITS;
2026
zone->length = blkz->len << BDRV_SECTOR_BITS;
2027
zone->wp = blkz->wp << BDRV_SECTOR_BITS;
2029
#ifdef HAVE_BLK_ZONE_REP_CAPACITY
2030
zone->cap = blkz->capacity << BDRV_SECTOR_BITS;
2032
zone->cap = blkz->len << BDRV_SECTOR_BITS;
2035
switch (blkz->type) {
2036
case BLK_ZONE_TYPE_SEQWRITE_REQ:
2037
zone->type = BLK_ZT_SWR;
2039
case BLK_ZONE_TYPE_SEQWRITE_PREF:
2040
zone->type = BLK_ZT_SWP;
2042
case BLK_ZONE_TYPE_CONVENTIONAL:
2043
zone->type = BLK_ZT_CONV;
2046
error_report("Unsupported zone type: 0x%x", blkz->type);
2050
switch (blkz->cond) {
2051
case BLK_ZONE_COND_NOT_WP:
2052
zone->state = BLK_ZS_NOT_WP;
2054
case BLK_ZONE_COND_EMPTY:
2055
zone->state = BLK_ZS_EMPTY;
2057
case BLK_ZONE_COND_IMP_OPEN:
2058
zone->state = BLK_ZS_IOPEN;
2060
case BLK_ZONE_COND_EXP_OPEN:
2061
zone->state = BLK_ZS_EOPEN;
2063
case BLK_ZONE_COND_CLOSED:
2064
zone->state = BLK_ZS_CLOSED;
2066
case BLK_ZONE_COND_READONLY:
2067
zone->state = BLK_ZS_RDONLY;
2069
case BLK_ZONE_COND_FULL:
2070
zone->state = BLK_ZS_FULL;
2072
case BLK_ZONE_COND_OFFLINE:
2073
zone->state = BLK_ZS_OFFLINE;
2076
error_report("Unsupported zone state: 0x%x", blkz->cond);
2083
#if defined(CONFIG_BLKZONED)
2084
static int handle_aiocb_zone_report(void *opaque)
2086
RawPosixAIOData *aiocb = opaque;
2087
int fd = aiocb->aio_fildes;
2088
unsigned int *nr_zones = aiocb->zone_report.nr_zones;
2089
BlockZoneDescriptor *zones = aiocb->zone_report.zones;
2090
/* zoned block devices use 512-byte sectors */
2091
uint64_t sector = aiocb->aio_offset / 512;
2093
struct blk_zone *blkz;
2097
unsigned int n = 0, i = 0;
2100
rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct blk_zone);
2101
g_autofree struct blk_zone_report *rep = NULL;
2102
rep = g_malloc(rep_size);
2104
blkz = (struct blk_zone *)(rep + 1);
2106
memset(rep, 0, rep_size);
2107
rep->sector = sector;
2108
rep->nr_zones = nrz - n;
2111
ret = ioctl(fd, BLKREPORTZONE, rep);
2112
} while (ret != 0 && errno == EINTR);
2114
error_report("%d: ioctl BLKREPORTZONE at %" PRId64 " failed %d",
2119
if (!rep->nr_zones) {
2123
for (i = 0; i < rep->nr_zones; i++, n++) {
2124
ret = parse_zone(&zones[n], &blkz[i]);
2129
/* The next report should start after the last zone reported */
2130
sector = blkz[i].start + blkz[i].len;
2139
#if defined(CONFIG_BLKZONED)
2140
static int handle_aiocb_zone_mgmt(void *opaque)
2142
RawPosixAIOData *aiocb = opaque;
2143
int fd = aiocb->aio_fildes;
2144
uint64_t sector = aiocb->aio_offset / 512;
2145
int64_t nr_sectors = aiocb->aio_nbytes / 512;
2146
struct blk_zone_range range;
2149
/* Execute the operation */
2150
range.sector = sector;
2151
range.nr_sectors = nr_sectors;
2153
ret = ioctl(fd, aiocb->zone_mgmt.op, &range);
2154
} while (ret != 0 && errno == EINTR);
2156
return ret < 0 ? -errno : ret;
2160
static int handle_aiocb_copy_range(void *opaque)
2162
RawPosixAIOData *aiocb = opaque;
2163
uint64_t bytes = aiocb->aio_nbytes;
2164
off_t in_off = aiocb->aio_offset;
2165
off_t out_off = aiocb->copy_range.aio_offset2;
2168
ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off,
2169
aiocb->copy_range.aio_fd2, &out_off,
2171
trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off,
2172
aiocb->copy_range.aio_fd2, out_off, bytes,
2175
/* No progress (e.g. when beyond EOF), let the caller fall back to
2194
static int handle_aiocb_discard(void *opaque)
2196
RawPosixAIOData *aiocb = opaque;
2198
BDRVRawState *s = aiocb->bs->opaque;
2200
if (!s->has_discard) {
2204
if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
2207
uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
2208
if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
2211
} while (errno == EINTR);
2213
ret = translate_err(-errno);
2216
#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
2217
ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
2218
aiocb->aio_offset, aiocb->aio_nbytes);
2219
ret = translate_err(ret);
2220
#elif defined(__APPLE__) && (__MACH__)
2221
fpunchhole_t fpunchhole;
2222
fpunchhole.fp_flags = 0;
2223
fpunchhole.reserved = 0;
2224
fpunchhole.fp_offset = aiocb->aio_offset;
2225
fpunchhole.fp_length = aiocb->aio_nbytes;
2226
if (fcntl(s->fd, F_PUNCHHOLE, &fpunchhole) == -1) {
2227
ret = errno == ENODEV ? -ENOTSUP : -errno;
2234
if (ret == -ENOTSUP) {
2235
s->has_discard = false;
2241
* Help alignment probing by allocating the first block.
2243
* When reading with direct I/O from unallocated area on Gluster backed by XFS,
2244
* reading succeeds regardless of request length. In this case we fallback to
2245
* safe alignment which is not optimal. Allocating the first block avoids this
2248
* fd may be opened with O_DIRECT, but we don't know the buffer alignment or
2249
* request alignment, so we use safe values.
2251
* Returns: 0 on success, -errno on failure. Since this is an optimization,
2252
* caller may ignore failures.
2254
static int allocate_first_block(int fd, size_t max_size)
2256
size_t write_size = (max_size < MAX_BLOCKSIZE)
2259
size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size());
2264
buf = qemu_memalign(max_align, write_size);
2265
memset(buf, 0, write_size);
2267
n = RETRY_ON_EINTR(pwrite(fd, buf, write_size, 0));
2269
ret = (n == -1) ? -errno : 0;
2275
static int handle_aiocb_truncate(void *opaque)
2277
RawPosixAIOData *aiocb = opaque;
2279
int64_t current_length = 0;
2282
int fd = aiocb->aio_fildes;
2283
int64_t offset = aiocb->aio_offset;
2284
PreallocMode prealloc = aiocb->truncate.prealloc;
2285
Error **errp = aiocb->truncate.errp;
2287
if (fstat(fd, &st) < 0) {
2289
error_setg_errno(errp, -result, "Could not stat file");
2293
current_length = st.st_size;
2294
if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
2295
error_setg(errp, "Cannot use preallocation for shrinking files");
2300
#ifdef CONFIG_POSIX_FALLOCATE
2301
case PREALLOC_MODE_FALLOC:
2303
* Truncating before posix_fallocate() makes it about twice slower on
2304
* file systems that do not support fallocate(), trying to check if a
2305
* block is allocated before allocating it, so don't do that here.
2307
if (offset != current_length) {
2308
result = -posix_fallocate(fd, current_length,
2309
offset - current_length);
2311
/* posix_fallocate() doesn't set errno. */
2312
error_setg_errno(errp, -result,
2313
"Could not preallocate new data");
2314
} else if (current_length == 0) {
2316
* posix_fallocate() uses fallocate() if the filesystem
2317
* supports it, or fallback to manually writing zeroes. If
2318
* fallocate() was used, unaligned reads from the fallocated
2319
* area in raw_probe_alignment() will succeed, hence we need to
2320
* allocate the first block.
2322
* Optimize future alignment probing; ignore failures.
2324
allocate_first_block(fd, offset);
2331
case PREALLOC_MODE_FULL:
2333
int64_t num = 0, left = offset - current_length;
2337
* Knowing the final size from the beginning could allow the file
2338
* system driver to do less allocations and possibly avoid
2339
* fragmentation of the file.
2341
if (ftruncate(fd, offset) != 0) {
2343
error_setg_errno(errp, -result, "Could not resize file");
2347
buf = g_malloc0(65536);
2349
seek_result = lseek(fd, current_length, SEEK_SET);
2350
if (seek_result < 0) {
2352
error_setg_errno(errp, -result,
2353
"Failed to seek to the old end of file");
2358
num = MIN(left, 65536);
2359
result = write(fd, buf, num);
2361
if (errno == EINTR) {
2365
error_setg_errno(errp, -result,
2366
"Could not write zeros for preallocation");
2375
error_setg_errno(errp, -result,
2376
"Could not flush file to disk");
2382
case PREALLOC_MODE_OFF:
2383
if (ftruncate(fd, offset) != 0) {
2385
error_setg_errno(errp, -result, "Could not resize file");
2386
} else if (current_length == 0 && offset > current_length) {
2387
/* Optimize future alignment probing; ignore failures. */
2388
allocate_first_block(fd, offset);
2393
error_setg(errp, "Unsupported preallocation mode: %s",
2394
PreallocMode_str(prealloc));
2400
if (ftruncate(fd, current_length) < 0) {
2401
error_report("Failed to restore old file length: %s",
2410
static int coroutine_fn raw_thread_pool_submit(ThreadPoolFunc func, void *arg)
2412
return thread_pool_submit_co(func, arg);
2416
* Check if all memory in this vector is sector aligned.
2418
static bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2421
size_t alignment = bdrv_min_mem_align(bs);
2422
size_t len = bs->bl.request_alignment;
2425
for (i = 0; i < qiov->niov; i++) {
2426
if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2429
if (qiov->iov[i].iov_len % len) {
2437
#ifdef CONFIG_LINUX_IO_URING
2438
static inline bool raw_check_linux_io_uring(BDRVRawState *s)
2440
Error *local_err = NULL;
2443
if (!s->use_linux_io_uring) {
2447
ctx = qemu_get_current_aio_context();
2448
if (unlikely(!aio_setup_linux_io_uring(ctx, &local_err))) {
2449
error_reportf_err(local_err, "Unable to use linux io_uring, "
2450
"falling back to thread pool: ");
2451
s->use_linux_io_uring = false;
2458
#ifdef CONFIG_LINUX_AIO
2459
static inline bool raw_check_linux_aio(BDRVRawState *s)
2461
Error *local_err = NULL;
2464
if (!s->use_linux_aio) {
2468
ctx = qemu_get_current_aio_context();
2469
if (unlikely(!aio_setup_linux_aio(ctx, &local_err))) {
2470
error_reportf_err(local_err, "Unable to use Linux AIO, "
2471
"falling back to thread pool: ");
2472
s->use_linux_aio = false;
2479
static int coroutine_fn raw_co_prw(BlockDriverState *bs, int64_t *offset_ptr,
2480
uint64_t bytes, QEMUIOVector *qiov, int type)
2482
BDRVRawState *s = bs->opaque;
2483
RawPosixAIOData acb;
2485
uint64_t offset = *offset_ptr;
2487
if (fd_open(bs) < 0)
2489
#if defined(CONFIG_BLKZONED)
2490
if ((type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) &&
2491
bs->bl.zoned != BLK_Z_NONE) {
2492
qemu_co_mutex_lock(&bs->wps->colock);
2493
if (type & QEMU_AIO_ZONE_APPEND) {
2494
int index = offset / bs->bl.zone_size;
2495
offset = bs->wps->wp[index];
2501
* When using O_DIRECT, the request must be aligned to be able to use
2502
* either libaio or io_uring interface. If not fail back to regular thread
2503
* pool read/write code which emulates this for us if we
2504
* set QEMU_AIO_MISALIGNED.
2506
if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) {
2507
type |= QEMU_AIO_MISALIGNED;
2508
#ifdef CONFIG_LINUX_IO_URING
2509
} else if (raw_check_linux_io_uring(s)) {
2510
assert(qiov->size == bytes);
2511
ret = luring_co_submit(bs, s->fd, offset, qiov, type);
2514
#ifdef CONFIG_LINUX_AIO
2515
} else if (raw_check_linux_aio(s)) {
2516
assert(qiov->size == bytes);
2517
ret = laio_co_submit(s->fd, offset, qiov, type,
2523
acb = (RawPosixAIOData) {
2525
.aio_fildes = s->fd,
2527
.aio_offset = offset,
2528
.aio_nbytes = bytes,
2535
assert(qiov->size == bytes);
2536
ret = raw_thread_pool_submit(handle_aiocb_rw, &acb);
2537
goto out; /* Avoid the compiler err of unused label */
2540
#if defined(CONFIG_BLKZONED)
2541
if ((type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) &&
2542
bs->bl.zoned != BLK_Z_NONE) {
2543
BlockZoneWps *wps = bs->wps;
2545
uint64_t *wp = &wps->wp[offset / bs->bl.zone_size];
2546
if (!BDRV_ZT_IS_CONV(*wp)) {
2547
if (type & QEMU_AIO_ZONE_APPEND) {
2549
trace_zbd_zone_append_complete(bs, *offset_ptr
2550
>> BDRV_SECTOR_BITS);
2552
/* Advance the wp if needed */
2553
if (offset + bytes > *wp) {
2554
*wp = offset + bytes;
2559
* write and append write are not allowed to cross zone boundaries
2561
update_zones_wp(bs, s->fd, offset, 1);
2564
qemu_co_mutex_unlock(&wps->colock);
2570
static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset,
2571
int64_t bytes, QEMUIOVector *qiov,
2572
BdrvRequestFlags flags)
2574
return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_READ);
2577
static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset,
2578
int64_t bytes, QEMUIOVector *qiov,
2579
BdrvRequestFlags flags)
2581
return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_WRITE);
2584
static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs)
2586
BDRVRawState *s = bs->opaque;
2587
RawPosixAIOData acb;
2595
acb = (RawPosixAIOData) {
2597
.aio_fildes = s->fd,
2598
.aio_type = QEMU_AIO_FLUSH,
2601
#ifdef CONFIG_LINUX_IO_URING
2602
if (raw_check_linux_io_uring(s)) {
2603
return luring_co_submit(bs, s->fd, 0, NULL, QEMU_AIO_FLUSH);
2606
#ifdef CONFIG_LINUX_AIO
2607
if (s->has_laio_fdsync && raw_check_linux_aio(s)) {
2608
return laio_co_submit(s->fd, 0, NULL, QEMU_AIO_FLUSH, 0);
2611
return raw_thread_pool_submit(handle_aiocb_flush, &acb);
2614
static void raw_close(BlockDriverState *bs)
2616
BDRVRawState *s = bs->opaque;
2619
#if defined(CONFIG_BLKZONED)
2628
* Truncates the given regular file @fd to @offset and, when growing, fills the
2629
* new space according to @prealloc.
2631
* Returns: 0 on success, -errno on failure.
2633
static int coroutine_fn
2634
raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset,
2635
PreallocMode prealloc, Error **errp)
2637
RawPosixAIOData acb;
2639
acb = (RawPosixAIOData) {
2642
.aio_type = QEMU_AIO_TRUNCATE,
2643
.aio_offset = offset,
2645
.prealloc = prealloc,
2650
return raw_thread_pool_submit(handle_aiocb_truncate, &acb);
2653
static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,
2654
bool exact, PreallocMode prealloc,
2655
BdrvRequestFlags flags, Error **errp)
2657
BDRVRawState *s = bs->opaque;
2661
if (fstat(s->fd, &st)) {
2663
error_setg_errno(errp, -ret, "Failed to fstat() the file");
2667
if (S_ISREG(st.st_mode)) {
2668
/* Always resizes to the exact @offset */
2669
return raw_regular_truncate(bs, s->fd, offset, prealloc, errp);
2672
if (prealloc != PREALLOC_MODE_OFF) {
2673
error_setg(errp, "Preallocation mode '%s' unsupported for this "
2674
"non-regular file", PreallocMode_str(prealloc));
2678
if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2679
int64_t cur_length = raw_getlength(bs);
2681
if (offset != cur_length && exact) {
2682
error_setg(errp, "Cannot resize device files");
2684
} else if (offset > cur_length) {
2685
error_setg(errp, "Cannot grow device files");
2689
error_setg(errp, "Resizing this file is not supported");
2697
static int64_t raw_getlength(BlockDriverState *bs)
2699
BDRVRawState *s = bs->opaque;
2705
if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2706
struct disklabel dl;
2708
if (ioctl(fd, DIOCGDINFO, &dl))
2710
return (uint64_t)dl.d_secsize *
2711
dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2715
#elif defined(__NetBSD__)
2716
static int64_t raw_getlength(BlockDriverState *bs)
2718
BDRVRawState *s = bs->opaque;
2724
if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2725
struct dkwedge_info dkw;
2727
if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
2728
return dkw.dkw_size * 512;
2730
struct disklabel dl;
2732
if (ioctl(fd, DIOCGDINFO, &dl))
2734
return (uint64_t)dl.d_secsize *
2735
dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2740
#elif defined(__sun__)
2741
static int64_t raw_getlength(BlockDriverState *bs)
2743
BDRVRawState *s = bs->opaque;
2744
struct dk_minfo minfo;
2754
* Use the DKIOCGMEDIAINFO ioctl to read the size.
2756
ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
2758
return minfo.dki_lbsize * minfo.dki_capacity;
2762
* There are reports that lseek on some devices fails, but
2763
* irc discussion said that contingency on contingency was overkill.
2765
size = lseek(s->fd, 0, SEEK_END);
2771
#elif defined(CONFIG_BSD)
2772
static int64_t raw_getlength(BlockDriverState *bs)
2774
BDRVRawState *s = bs->opaque;
2778
#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2787
#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2790
if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
2792
#ifdef DIOCGMEDIASIZE
2793
if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) {
2800
if (ioctl(fd, DIOCGPART, &pi) == 0) {
2801
size = pi.media_size;
2805
#if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE)
2807
uint64_t sectors = 0;
2808
uint32_t sector_size = 0;
2810
if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
2811
&& ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
2812
size = sectors * sector_size;
2817
size = lseek(fd, 0LL, SEEK_END);
2822
#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2825
/* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2826
if (size == 2048LL * (unsigned)-1)
2828
/* XXX no disc? maybe we need to reopen... */
2829
if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
2836
size = lseek(fd, 0, SEEK_END);
2844
static int64_t raw_getlength(BlockDriverState *bs)
2846
BDRVRawState *s = bs->opaque;
2855
size = lseek(s->fd, 0, SEEK_END);
2863
static int64_t coroutine_fn raw_co_getlength(BlockDriverState *bs)
2865
return raw_getlength(bs);
2868
static int64_t coroutine_fn raw_co_get_allocated_file_size(BlockDriverState *bs)
2871
BDRVRawState *s = bs->opaque;
2873
if (fstat(s->fd, &st) < 0) {
2876
return (int64_t)st.st_blocks * 512;
2879
static int coroutine_fn
2880
raw_co_create(BlockdevCreateOptions *options, Error **errp)
2882
BlockdevCreateOptionsFile *file_opts;
2883
Error *local_err = NULL;
2885
uint64_t perm, shared;
2888
/* Validate options and set default values */
2889
assert(options->driver == BLOCKDEV_DRIVER_FILE);
2890
file_opts = &options->u.file;
2892
if (!file_opts->has_nocow) {
2893
file_opts->nocow = false;
2895
if (!file_opts->has_preallocation) {
2896
file_opts->preallocation = PREALLOC_MODE_OFF;
2898
if (!file_opts->has_extent_size_hint) {
2899
file_opts->extent_size_hint = 1 * MiB;
2901
if (file_opts->extent_size_hint > UINT32_MAX) {
2903
error_setg(errp, "Extent size hint is too large");
2908
fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp);
2914
/* Take permissions: We want to discard everything, so we need
2915
* BLK_PERM_WRITE; and truncation to the desired size requires
2917
* On the other hand, we cannot share the RESIZE permission
2918
* because we promise that after this function, the file has the
2919
* size given in the options. If someone else were to resize it
2920
* concurrently, we could not guarantee that.
2921
* Note that after this function, we can no longer guarantee that
2922
* the file is not touched by a third party, so it may be resized
2924
perm = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2925
shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
2927
/* Step one: Take locks */
2928
result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp);
2933
/* Step two: Check that nobody else has taken conflicting locks */
2934
result = raw_check_lock_bytes(fd, perm, shared, errp);
2936
error_append_hint(errp,
2937
"Is another process using the image [%s]?\n",
2938
file_opts->filename);
2942
/* Clear the file by truncating it to 0 */
2943
result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp);
2948
if (file_opts->nocow) {
2950
/* Set NOCOW flag to solve performance issue on fs like btrfs.
2951
* This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2952
* will be ignored since any failure of this operation should not
2953
* block the left work.
2956
if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2957
attr |= FS_NOCOW_FL;
2958
ioctl(fd, FS_IOC_SETFLAGS, &attr);
2962
#ifdef FS_IOC_FSSETXATTR
2964
* Try to set the extent size hint. Failure is not fatal, and a warning is
2965
* only printed if the option was explicitly specified.
2968
struct fsxattr attr;
2969
result = ioctl(fd, FS_IOC_FSGETXATTR, &attr);
2971
attr.fsx_xflags |= FS_XFLAG_EXTSIZE;
2972
attr.fsx_extsize = file_opts->extent_size_hint;
2973
result = ioctl(fd, FS_IOC_FSSETXATTR, &attr);
2975
if (result < 0 && file_opts->has_extent_size_hint &&
2976
file_opts->extent_size_hint)
2978
warn_report("Failed to set extent size hint: %s",
2984
/* Resize and potentially preallocate the file to the desired
2986
result = raw_regular_truncate(NULL, fd, file_opts->size,
2987
file_opts->preallocation, errp);
2993
raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err);
2995
/* The above call should not fail, and if it does, that does
2996
* not mean the whole creation operation has failed. So
2997
* report it the user for their convenience, but do not report
2998
* it to the caller. */
2999
warn_report_err(local_err);
3003
if (qemu_close(fd) != 0 && result == 0) {
3005
error_setg_errno(errp, -result, "Could not close the new file");
3011
static int coroutine_fn GRAPH_RDLOCK
3012
raw_co_create_opts(BlockDriver *drv, const char *filename,
3013
QemuOpts *opts, Error **errp)
3015
BlockdevCreateOptions options;
3016
int64_t total_size = 0;
3017
int64_t extent_size_hint = 0;
3018
bool has_extent_size_hint = false;
3020
PreallocMode prealloc;
3022
Error *local_err = NULL;
3024
/* Skip file: protocol prefix */
3025
strstart(filename, "file:", &filename);
3027
/* Read out options */
3028
total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
3030
if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) {
3031
has_extent_size_hint = true;
3033
qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1);
3035
nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
3036
buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
3037
prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
3038
PREALLOC_MODE_OFF, &local_err);
3041
error_propagate(errp, local_err);
3045
options = (BlockdevCreateOptions) {
3046
.driver = BLOCKDEV_DRIVER_FILE,
3048
.filename = (char *) filename,
3050
.has_preallocation = true,
3051
.preallocation = prealloc,
3054
.has_extent_size_hint = has_extent_size_hint,
3055
.extent_size_hint = extent_size_hint,
3058
return raw_co_create(&options, errp);
3061
static int coroutine_fn raw_co_delete_file(BlockDriverState *bs,
3067
if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) {
3068
error_setg_errno(errp, ENOENT, "%s is not a regular file",
3073
ret = unlink(bs->filename);
3076
error_setg_errno(errp, -ret, "Error when deleting file %s",
3084
* Find allocation range in @bs around offset @start.
3085
* May change underlying file descriptor's file offset.
3086
* If @start is not in a hole, store @start in @data, and the
3087
* beginning of the next hole in @hole, and return 0.
3088
* If @start is in a non-trailing hole, store @start in @hole and the
3089
* beginning of the next non-hole in @data, and return 0.
3090
* If @start is in a trailing hole or beyond EOF, return -ENXIO.
3091
* If we can't find out, return a negative errno other than -ENXIO.
3093
static int find_allocation(BlockDriverState *bs, off_t start,
3094
off_t *data, off_t *hole)
3096
#if defined SEEK_HOLE && defined SEEK_DATA
3097
BDRVRawState *s = bs->opaque;
3102
* D1. offs == start: start is in data
3103
* D2. offs > start: start is in a hole, next data at offs
3104
* D3. offs < 0, errno = ENXIO: either start is in a trailing hole
3105
* or start is beyond EOF
3106
* If the latter happens, the file has been truncated behind
3107
* our back since we opened it. All bets are off then.
3108
* Treating like a trailing hole is simplest.
3109
* D4. offs < 0, errno != ENXIO: we learned nothing
3111
offs = lseek(s->fd, start, SEEK_DATA);
3113
return -errno; /* D3 or D4 */
3117
/* This is not a valid return by lseek(). We are safe to just return
3118
* -EIO in this case, and we'll treat it like D4. */
3123
/* D2: in hole, next data at offs */
3129
/* D1: in data, end not yet known */
3133
* H1. offs == start: start is in a hole
3134
* If this happens here, a hole has been dug behind our back
3135
* since the previous lseek().
3136
* H2. offs > start: either start is in data, next hole at offs,
3137
* or start is in trailing hole, EOF at offs
3138
* Linux treats trailing holes like any other hole: offs ==
3139
* start. Solaris seeks to EOF instead: offs > start (blech).
3140
* If that happens here, a hole has been dug behind our back
3141
* since the previous lseek().
3142
* H3. offs < 0, errno = ENXIO: start is beyond EOF
3143
* If this happens, the file has been truncated behind our
3144
* back since we opened it. Treat it like a trailing hole.
3145
* H4. offs < 0, errno != ENXIO: we learned nothing
3146
* Pretend we know nothing at all, i.e. "forget" about D1.
3148
offs = lseek(s->fd, start, SEEK_HOLE);
3150
return -errno; /* D1 and (H3 or H4) */
3154
/* This is not a valid return by lseek(). We are safe to just return
3155
* -EIO in this case, and we'll treat it like H4. */
3161
* D1 and H2: either in data, next hole at offs, or it was in
3162
* data but is now in a trailing hole. In the latter case,
3163
* all bets are off. Treating it as if it there was data all
3164
* the way to EOF is safe, so simply do that.
3179
* Returns the allocation status of the specified offset.
3181
* The block layer guarantees 'offset' and 'bytes' are within bounds.
3183
* 'pnum' is set to the number of bytes (including and immediately following
3184
* the specified offset) that are known to be in the same
3185
* allocated/unallocated state.
3187
* 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may
3190
static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
3193
int64_t bytes, int64_t *pnum,
3195
BlockDriverState **file)
3197
off_t data = 0, hole = 0;
3200
assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
3211
return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
3214
ret = find_allocation(bs, offset, &data, &hole);
3215
if (ret == -ENXIO) {
3218
ret = BDRV_BLOCK_ZERO;
3219
} else if (ret < 0) {
3220
/* No info available, so pretend there are no holes */
3222
ret = BDRV_BLOCK_DATA;
3223
} else if (data == offset) {
3224
/* On a data extent, compute bytes to the end of the extent,
3225
* possibly including a partial sector at EOF. */
3226
*pnum = hole - offset;
3229
* We are not allowed to return partial sectors, though, so
3230
* round up if necessary.
3232
if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) {
3233
int64_t file_length = raw_getlength(bs);
3234
if (file_length > 0) {
3235
/* Ignore errors, this is just a safeguard */
3236
assert(hole == file_length);
3238
*pnum = ROUND_UP(*pnum, bs->bl.request_alignment);
3241
ret = BDRV_BLOCK_DATA;
3243
/* On a hole, compute bytes to the beginning of the next extent. */
3244
assert(hole == offset);
3245
*pnum = data - offset;
3246
ret = BDRV_BLOCK_ZERO;
3250
return ret | BDRV_BLOCK_OFFSET_VALID;
3253
#if defined(__linux__)
3254
/* Verify that the file is not in the page cache */
3255
static void check_cache_dropped(BlockDriverState *bs, Error **errp)
3257
const size_t window_size = 128 * 1024 * 1024;
3258
BDRVRawState *s = bs->opaque;
3259
void *window = NULL;
3266
/* mincore(2) page status information requires 1 byte per page */
3267
page_size = sysconf(_SC_PAGESIZE);
3268
vec = g_malloc(DIV_ROUND_UP(window_size, page_size));
3270
end = raw_getlength(bs);
3272
for (offset = 0; offset < end; offset += window_size) {
3279
/* Unmap previous window if size has changed */
3280
new_length = MIN(end - offset, window_size);
3281
if (new_length != length) {
3282
munmap(window, length);
3287
new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE,
3289
if (new_window == MAP_FAILED) {
3290
error_setg_errno(errp, errno, "mmap failed");
3294
window = new_window;
3295
length = new_length;
3297
ret = mincore(window, length, vec);
3299
error_setg_errno(errp, errno, "mincore failed");
3303
vec_end = DIV_ROUND_UP(length, page_size);
3304
for (i = 0; i < vec_end; i++) {
3310
error_setg(errp, "page cache still in use!");
3316
munmap(window, length);
3321
#endif /* __linux__ */
3323
static void coroutine_fn GRAPH_RDLOCK
3324
raw_co_invalidate_cache(BlockDriverState *bs, Error **errp)
3326
BDRVRawState *s = bs->opaque;
3331
error_setg_errno(errp, -ret, "The file descriptor is not open");
3335
if (!s->drop_cache) {
3339
if (s->open_flags & O_DIRECT) {
3340
return; /* No host kernel page cache */
3343
#if defined(__linux__)
3344
/* This sets the scene for the next syscall... */
3345
ret = bdrv_co_flush(bs);
3347
error_setg_errno(errp, -ret, "flush failed");
3351
/* Linux does not invalidate pages that are dirty, locked, or mmapped by a
3352
* process. These limitations are okay because we just fsynced the file,
3353
* we don't use mmap, and the file should not be in use by other processes.
3355
ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED);
3356
if (ret != 0) { /* the return value is a positive errno */
3357
error_setg_errno(errp, ret, "fadvise failed");
3361
if (s->check_cache_dropped) {
3362
check_cache_dropped(bs, errp);
3364
#else /* __linux__ */
3365
/* Do nothing. Live migration to a remote host with cache.direct=off is
3366
* unsupported on other host operating systems. Cache consistency issues
3367
* may occur but no error is reported here, partly because that's the
3368
* historical behavior and partly because it's hard to differentiate valid
3369
* configurations that should not cause errors.
3371
#endif /* !__linux__ */
3374
static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret)
3377
s->stats.discard_nb_failed++;
3379
s->stats.discard_nb_ok++;
3380
s->stats.discard_bytes_ok += nbytes;
3385
* zone report - Get a zone block device's information in the form
3386
* of an array of zone descriptors.
3387
* zones is an array of zone descriptors to hold zone information on reply;
3388
* offset can be any byte within the entire size of the device;
3389
* nr_zones is the maximum number of sectors the command should operate on.
3391
#if defined(CONFIG_BLKZONED)
3392
static int coroutine_fn raw_co_zone_report(BlockDriverState *bs, int64_t offset,
3393
unsigned int *nr_zones,
3394
BlockZoneDescriptor *zones) {
3395
BDRVRawState *s = bs->opaque;
3396
RawPosixAIOData acb = (RawPosixAIOData) {
3398
.aio_fildes = s->fd,
3399
.aio_type = QEMU_AIO_ZONE_REPORT,
3400
.aio_offset = offset,
3402
.nr_zones = nr_zones,
3407
trace_zbd_zone_report(bs, *nr_zones, offset >> BDRV_SECTOR_BITS);
3408
return raw_thread_pool_submit(handle_aiocb_zone_report, &acb);
3413
* zone management operations - Execute an operation on a zone
3415
#if defined(CONFIG_BLKZONED)
3416
static int coroutine_fn raw_co_zone_mgmt(BlockDriverState *bs, BlockZoneOp op,
3417
int64_t offset, int64_t len) {
3418
BDRVRawState *s = bs->opaque;
3419
RawPosixAIOData acb;
3420
int64_t zone_size, zone_size_mask;
3421
const char *op_name;
3424
BlockZoneWps *wps = bs->wps;
3425
int64_t capacity = bs->total_sectors << BDRV_SECTOR_BITS;
3427
zone_size = bs->bl.zone_size;
3428
zone_size_mask = zone_size - 1;
3429
if (offset & zone_size_mask) {
3430
error_report("sector offset %" PRId64 " is not aligned to zone size "
3431
"%" PRId64 "", offset / 512, zone_size / 512);
3435
if (((offset + len) < capacity && len & zone_size_mask) ||
3436
offset + len > capacity) {
3437
error_report("number of sectors %" PRId64 " is not aligned to zone size"
3438
" %" PRId64 "", len / 512, zone_size / 512);
3442
uint32_t i = offset / bs->bl.zone_size;
3443
uint32_t nrz = len / bs->bl.zone_size;
3444
uint64_t *wp = &wps->wp[i];
3445
if (BDRV_ZT_IS_CONV(*wp) && len != capacity) {
3446
error_report("zone mgmt operations are not allowed for conventional zones");
3452
op_name = "BLKOPENZONE";
3456
op_name = "BLKCLOSEZONE";
3460
op_name = "BLKFINISHZONE";
3464
op_name = "BLKRESETZONE";
3468
error_report("Unsupported zone op: 0x%x", op);
3472
acb = (RawPosixAIOData) {
3474
.aio_fildes = s->fd,
3475
.aio_type = QEMU_AIO_ZONE_MGMT,
3476
.aio_offset = offset,
3483
trace_zbd_zone_mgmt(bs, op_name, offset >> BDRV_SECTOR_BITS,
3484
len >> BDRV_SECTOR_BITS);
3485
ret = raw_thread_pool_submit(handle_aiocb_zone_mgmt, &acb);
3487
update_zones_wp(bs, s->fd, offset, nrz);
3488
error_report("ioctl %s failed %d", op_name, ret);
3492
if (zo == BLKRESETZONE && len == capacity) {
3493
ret = get_zones_wp(bs, s->fd, 0, bs->bl.nr_zones, 1);
3495
error_report("reporting single wp failed");
3498
} else if (zo == BLKRESETZONE) {
3499
for (unsigned int j = 0; j < nrz; ++j) {
3500
wp[j] = offset + j * zone_size;
3502
} else if (zo == BLKFINISHZONE) {
3503
for (unsigned int j = 0; j < nrz; ++j) {
3504
/* The zoned device allows the last zone smaller that the
3506
wp[j] = MIN(offset + (j + 1) * zone_size, offset + len);
3514
#if defined(CONFIG_BLKZONED)
3515
static int coroutine_fn raw_co_zone_append(BlockDriverState *bs,
3518
BdrvRequestFlags flags) {
3520
int64_t zone_size_mask = bs->bl.zone_size - 1;
3521
int64_t iov_len = 0;
3524
if (*offset & zone_size_mask) {
3525
error_report("sector offset %" PRId64 " is not aligned to zone size "
3526
"%" PRId32 "", *offset / 512, bs->bl.zone_size / 512);
3530
int64_t wg = bs->bl.write_granularity;
3531
int64_t wg_mask = wg - 1;
3532
for (int i = 0; i < qiov->niov; i++) {
3533
iov_len = qiov->iov[i].iov_len;
3534
if (iov_len & wg_mask) {
3535
error_report("len of IOVector[%d] %" PRId64 " is not aligned to "
3536
"block size %" PRId64 "", i, iov_len, wg);
3542
trace_zbd_zone_append(bs, *offset >> BDRV_SECTOR_BITS);
3543
return raw_co_prw(bs, offset, len, qiov, QEMU_AIO_ZONE_APPEND);
3547
static coroutine_fn int
3548
raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes,
3551
BDRVRawState *s = bs->opaque;
3552
RawPosixAIOData acb;
3555
acb = (RawPosixAIOData) {
3557
.aio_fildes = s->fd,
3558
.aio_type = QEMU_AIO_DISCARD,
3559
.aio_offset = offset,
3560
.aio_nbytes = bytes,
3564
acb.aio_type |= QEMU_AIO_BLKDEV;
3567
ret = raw_thread_pool_submit(handle_aiocb_discard, &acb);
3568
raw_account_discard(s, bytes, ret);
3572
static coroutine_fn int
3573
raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
3575
return raw_do_pdiscard(bs, offset, bytes, false);
3578
static int coroutine_fn
3579
raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
3580
BdrvRequestFlags flags, bool blkdev)
3582
BDRVRawState *s = bs->opaque;
3583
RawPosixAIOData acb;
3584
ThreadPoolFunc *handler;
3586
#ifdef CONFIG_FALLOCATE
3587
if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3588
BdrvTrackedRequest *req;
3591
* This is a workaround for a bug in the Linux XFS driver,
3592
* where writes submitted through the AIO interface will be
3593
* discarded if they happen beyond a concurrently running
3594
* fallocate() that increases the file length (i.e., both the
3595
* write and the fallocate() happen beyond the EOF).
3597
* To work around it, we extend the tracked request for this
3598
* zero write until INT64_MAX (effectively infinity), and mark
3599
* it as serializing.
3601
* We have to enable this workaround for all filesystems and
3602
* AIO modes (not just XFS with aio=native), because for
3603
* remote filesystems we do not know the host configuration.
3606
req = bdrv_co_get_self_request(bs);
3608
assert(req->type == BDRV_TRACKED_WRITE);
3609
assert(req->offset <= offset);
3610
assert(req->offset + req->bytes >= offset + bytes);
3612
req->bytes = BDRV_MAX_LENGTH - req->offset;
3614
bdrv_check_request(req->offset, req->bytes, &error_abort);
3616
bdrv_make_request_serialising(req, bs->bl.request_alignment);
3620
acb = (RawPosixAIOData) {
3622
.aio_fildes = s->fd,
3623
.aio_type = QEMU_AIO_WRITE_ZEROES,
3624
.aio_offset = offset,
3625
.aio_nbytes = bytes,
3629
acb.aio_type |= QEMU_AIO_BLKDEV;
3631
if (flags & BDRV_REQ_NO_FALLBACK) {
3632
acb.aio_type |= QEMU_AIO_NO_FALLBACK;
3635
if (flags & BDRV_REQ_MAY_UNMAP) {
3636
acb.aio_type |= QEMU_AIO_DISCARD;
3637
handler = handle_aiocb_write_zeroes_unmap;
3639
handler = handle_aiocb_write_zeroes;
3642
return raw_thread_pool_submit(handler, &acb);
3645
static int coroutine_fn raw_co_pwrite_zeroes(
3646
BlockDriverState *bs, int64_t offset,
3647
int64_t bytes, BdrvRequestFlags flags)
3649
return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false);
3652
static int coroutine_fn
3653
raw_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3658
static ImageInfoSpecific *raw_get_specific_info(BlockDriverState *bs,
3661
ImageInfoSpecificFile *file_info = g_new0(ImageInfoSpecificFile, 1);
3662
ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
3664
*spec_info = (ImageInfoSpecific){
3665
.type = IMAGE_INFO_SPECIFIC_KIND_FILE,
3666
.u.file.data = file_info,
3669
#ifdef FS_IOC_FSGETXATTR
3671
BDRVRawState *s = bs->opaque;
3672
struct fsxattr attr;
3675
ret = ioctl(s->fd, FS_IOC_FSGETXATTR, &attr);
3676
if (!ret && attr.fsx_extsize != 0) {
3677
file_info->has_extent_size_hint = true;
3678
file_info->extent_size_hint = attr.fsx_extsize;
3686
static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs)
3688
BDRVRawState *s = bs->opaque;
3689
return (BlockStatsSpecificFile) {
3690
.discard_nb_ok = s->stats.discard_nb_ok,
3691
.discard_nb_failed = s->stats.discard_nb_failed,
3692
.discard_bytes_ok = s->stats.discard_bytes_ok,
3696
static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs)
3698
BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
3700
stats->driver = BLOCKDEV_DRIVER_FILE;
3701
stats->u.file = get_blockstats_specific_file(bs);
3706
#if defined(HAVE_HOST_BLOCK_DEVICE)
3707
static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs)
3709
BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
3711
stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE;
3712
stats->u.host_device = get_blockstats_specific_file(bs);
3716
#endif /* HAVE_HOST_BLOCK_DEVICE */
3718
static QemuOptsList raw_create_opts = {
3719
.name = "raw-create-opts",
3720
.head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
3723
.name = BLOCK_OPT_SIZE,
3724
.type = QEMU_OPT_SIZE,
3725
.help = "Virtual disk size"
3728
.name = BLOCK_OPT_NOCOW,
3729
.type = QEMU_OPT_BOOL,
3730
.help = "Turn off copy-on-write (valid only on btrfs)"
3733
.name = BLOCK_OPT_PREALLOC,
3734
.type = QEMU_OPT_STRING,
3735
.help = "Preallocation mode (allowed values: off"
3736
#ifdef CONFIG_POSIX_FALLOCATE
3742
.name = BLOCK_OPT_EXTENT_SIZE_HINT,
3743
.type = QEMU_OPT_SIZE,
3744
.help = "Extent size hint for the image file, 0 to disable"
3746
{ /* end of list */ }
3750
static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
3753
BDRVRawState *s = bs->opaque;
3754
int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags;
3758
/* We may need a new fd if auto-read-only switches the mode */
3759
ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, errp);
3762
} else if (ret != s->fd) {
3763
Error *local_err = NULL;
3766
* Fail already check_perm() if we can't get a working O_DIRECT
3767
* alignment with the new fd.
3769
raw_probe_alignment(bs, ret, &local_err);
3771
error_propagate(errp, local_err);
3775
s->perm_change_fd = ret;
3776
s->perm_change_flags = open_flags;
3779
/* Prepare permissions on old fd to avoid conflicts between old and new,
3780
* but keep everything locked that new will need. */
3781
ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
3786
/* Copy locks to the new fd */
3787
if (s->perm_change_fd && s->use_lock) {
3788
ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared,
3791
raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
3798
if (s->perm_change_fd) {
3799
qemu_close(s->perm_change_fd);
3801
s->perm_change_fd = 0;
3805
static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
3807
BDRVRawState *s = bs->opaque;
3809
/* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3810
* called after .bdrv_reopen_commit) */
3811
if (s->perm_change_fd && s->fd != s->perm_change_fd) {
3813
s->fd = s->perm_change_fd;
3814
s->open_flags = s->perm_change_flags;
3816
s->perm_change_fd = 0;
3818
raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
3820
s->shared_perm = shared;
3823
static void raw_abort_perm_update(BlockDriverState *bs)
3825
BDRVRawState *s = bs->opaque;
3827
/* For reopen, .bdrv_reopen_abort is called afterwards and will close
3828
* the file descriptor. */
3829
if (s->perm_change_fd) {
3830
qemu_close(s->perm_change_fd);
3832
s->perm_change_fd = 0;
3834
raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
3837
static int coroutine_fn GRAPH_RDLOCK raw_co_copy_range_from(
3838
BlockDriverState *bs, BdrvChild *src, int64_t src_offset,
3839
BdrvChild *dst, int64_t dst_offset, int64_t bytes,
3840
BdrvRequestFlags read_flags, BdrvRequestFlags write_flags)
3842
return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
3843
read_flags, write_flags);
3846
static int coroutine_fn GRAPH_RDLOCK
3847
raw_co_copy_range_to(BlockDriverState *bs,
3848
BdrvChild *src, int64_t src_offset,
3849
BdrvChild *dst, int64_t dst_offset,
3850
int64_t bytes, BdrvRequestFlags read_flags,
3851
BdrvRequestFlags write_flags)
3853
RawPosixAIOData acb;
3854
BDRVRawState *s = bs->opaque;
3855
BDRVRawState *src_s;
3857
assert(dst->bs == bs);
3858
if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) {
3862
src_s = src->bs->opaque;
3863
if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) {
3867
acb = (RawPosixAIOData) {
3869
.aio_type = QEMU_AIO_COPY_RANGE,
3870
.aio_fildes = src_s->fd,
3871
.aio_offset = src_offset,
3872
.aio_nbytes = bytes,
3875
.aio_offset2 = dst_offset,
3879
return raw_thread_pool_submit(handle_aiocb_copy_range, &acb);
3882
BlockDriver bdrv_file = {
3883
.format_name = "file",
3884
.protocol_name = "file",
3885
.instance_size = sizeof(BDRVRawState),
3886
.bdrv_needs_filename = true,
3887
.bdrv_probe = NULL, /* no probe for protocols */
3888
.bdrv_parse_filename = raw_parse_filename,
3889
.bdrv_open = raw_open,
3890
.bdrv_reopen_prepare = raw_reopen_prepare,
3891
.bdrv_reopen_commit = raw_reopen_commit,
3892
.bdrv_reopen_abort = raw_reopen_abort,
3893
.bdrv_close = raw_close,
3894
.bdrv_co_create = raw_co_create,
3895
.bdrv_co_create_opts = raw_co_create_opts,
3896
.bdrv_has_zero_init = bdrv_has_zero_init_1,
3897
.bdrv_co_block_status = raw_co_block_status,
3898
.bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3899
.bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
3900
.bdrv_co_delete_file = raw_co_delete_file,
3902
.bdrv_co_preadv = raw_co_preadv,
3903
.bdrv_co_pwritev = raw_co_pwritev,
3904
.bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3905
.bdrv_co_pdiscard = raw_co_pdiscard,
3906
.bdrv_co_copy_range_from = raw_co_copy_range_from,
3907
.bdrv_co_copy_range_to = raw_co_copy_range_to,
3908
.bdrv_refresh_limits = raw_refresh_limits,
3910
.bdrv_co_truncate = raw_co_truncate,
3911
.bdrv_co_getlength = raw_co_getlength,
3912
.bdrv_co_get_info = raw_co_get_info,
3913
.bdrv_get_specific_info = raw_get_specific_info,
3914
.bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
3915
.bdrv_get_specific_stats = raw_get_specific_stats,
3916
.bdrv_check_perm = raw_check_perm,
3917
.bdrv_set_perm = raw_set_perm,
3918
.bdrv_abort_perm_update = raw_abort_perm_update,
3919
.create_opts = &raw_create_opts,
3920
.mutable_opts = mutable_opts,
3923
/***********************************************/
3926
#if defined(HAVE_HOST_BLOCK_DEVICE)
3928
#if defined(__APPLE__) && defined(__MACH__)
3929
static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3930
CFIndex maxPathSize, int flags);
3932
static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
3934
kern_return_t kernResult = KERN_FAILURE;
3935
mach_port_t mainPort;
3936
CFMutableDictionaryRef classesToMatch;
3937
const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
3938
char *mediaType = NULL;
3940
kernResult = IOMainPort(MACH_PORT_NULL, &mainPort);
3941
if ( KERN_SUCCESS != kernResult ) {
3942
printf("IOMainPort returned %d\n", kernResult);
3946
for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
3947
classesToMatch = IOServiceMatching(matching_array[index]);
3948
if (classesToMatch == NULL) {
3949
error_report("IOServiceMatching returned NULL for %s",
3950
matching_array[index]);
3953
CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
3955
kernResult = IOServiceGetMatchingServices(mainPort, classesToMatch,
3957
if (kernResult != KERN_SUCCESS) {
3958
error_report("Note: IOServiceGetMatchingServices returned %d",
3963
/* If a match was found, leave the loop */
3964
if (*mediaIterator != 0) {
3965
trace_file_FindEjectableOpticalMedia(matching_array[index]);
3966
mediaType = g_strdup(matching_array[index]);
3973
kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3974
CFIndex maxPathSize, int flags)
3976
io_object_t nextMedia;
3977
kern_return_t kernResult = KERN_FAILURE;
3979
nextMedia = IOIteratorNext( mediaIterator );
3982
CFTypeRef bsdPathAsCFString;
3983
bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
3984
if ( bsdPathAsCFString ) {
3985
size_t devPathLength;
3986
strcpy( bsdPath, _PATH_DEV );
3987
if (flags & BDRV_O_NOCACHE) {
3988
strcat(bsdPath, "r");
3990
devPathLength = strlen( bsdPath );
3991
if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
3992
kernResult = KERN_SUCCESS;
3994
CFRelease( bsdPathAsCFString );
3996
IOObjectRelease( nextMedia );
4002
/* Sets up a real cdrom for use in QEMU */
4003
static bool setup_cdrom(char *bsd_path, Error **errp)
4005
int index, num_of_test_partitions = 2, fd;
4006
char test_partition[MAXPATHLEN];
4007
bool partition_found = false;
4009
/* look for a working partition */
4010
for (index = 0; index < num_of_test_partitions; index++) {
4011
snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
4013
fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL);
4015
partition_found = true;
4021
/* if a working partition on the device was not found */
4022
if (partition_found == false) {
4023
error_setg(errp, "Failed to find a working partition on disc");
4025
trace_file_setup_cdrom(test_partition);
4026
pstrcpy(bsd_path, MAXPATHLEN, test_partition);
4028
return partition_found;
4031
/* Prints directions on mounting and unmounting a device */
4032
static void print_unmounting_directions(const char *file_name)
4034
error_report("If device %s is mounted on the desktop, unmount"
4035
" it first before using it in QEMU", file_name);
4036
error_report("Command to unmount device: diskutil unmountDisk %s",
4038
error_report("Command to mount device: diskutil mountDisk %s", file_name);
4041
#endif /* defined(__APPLE__) && defined(__MACH__) */
4043
static int hdev_probe_device(const char *filename)
4047
/* allow a dedicated CD-ROM driver to match with a higher priority */
4048
if (strstart(filename, "/dev/cdrom", NULL))
4051
if (stat(filename, &st) >= 0 &&
4052
(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
4059
static void hdev_parse_filename(const char *filename, QDict *options,
4062
bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
4065
static bool hdev_is_sg(BlockDriverState *bs)
4068
#if defined(__linux__)
4070
BDRVRawState *s = bs->opaque;
4072
struct sg_scsi_id scsiid;
4076
if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
4080
ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
4085
ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
4087
trace_file_hdev_is_sg(scsiid.scsi_type, sg_version);
4096
static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
4099
BDRVRawState *s = bs->opaque;
4102
#if defined(__APPLE__) && defined(__MACH__)
4104
* Caution: while qdict_get_str() is fine, getting non-string types
4105
* would require more care. When @options come from -blockdev or
4106
* blockdev_add, its members are typed according to the QAPI
4107
* schema, but when they come from -drive, they're all QString.
4109
const char *filename = qdict_get_str(options, "filename");
4110
char bsd_path[MAXPATHLEN] = "";
4111
bool error_occurred = false;
4113
/* If using a real cdrom */
4114
if (strcmp(filename, "/dev/cdrom") == 0) {
4115
char *mediaType = NULL;
4116
kern_return_t ret_val;
4117
io_iterator_t mediaIterator = 0;
4119
mediaType = FindEjectableOpticalMedia(&mediaIterator);
4120
if (mediaType == NULL) {
4121
error_setg(errp, "Please make sure your CD/DVD is in the optical"
4123
error_occurred = true;
4124
goto hdev_open_Mac_error;
4127
ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
4128
if (ret_val != KERN_SUCCESS) {
4129
error_setg(errp, "Could not get BSD path for optical drive");
4130
error_occurred = true;
4131
goto hdev_open_Mac_error;
4134
/* If a real optical drive was not found */
4135
if (bsd_path[0] == '\0') {
4136
error_setg(errp, "Failed to obtain bsd path for optical drive");
4137
error_occurred = true;
4138
goto hdev_open_Mac_error;
4141
/* If using a cdrom disc and finding a partition on the disc failed */
4142
if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
4143
setup_cdrom(bsd_path, errp) == false) {
4144
print_unmounting_directions(bsd_path);
4145
error_occurred = true;
4146
goto hdev_open_Mac_error;
4149
qdict_put_str(options, "filename", bsd_path);
4153
if (mediaIterator) {
4154
IOObjectRelease(mediaIterator);
4156
if (error_occurred) {
4160
#endif /* defined(__APPLE__) && defined(__MACH__) */
4162
s->type = FTYPE_FILE;
4164
ret = raw_open_common(bs, options, flags, 0, true, errp);
4166
#if defined(__APPLE__) && defined(__MACH__)
4168
filename = bsd_path;
4170
/* if a physical device experienced an error while being opened */
4171
if (strncmp(filename, "/dev/", 5) == 0) {
4172
print_unmounting_directions(filename);
4174
#endif /* defined(__APPLE__) && defined(__MACH__) */
4178
/* Since this does ioctl the device must be already opened */
4179
bs->sg = hdev_is_sg(bs);
4184
#if defined(__linux__)
4185
static int coroutine_fn
4186
hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
4188
BDRVRawState *s = bs->opaque;
4189
RawPosixAIOData acb;
4197
if (req == SG_IO && s->pr_mgr) {
4198
struct sg_io_hdr *io_hdr = buf;
4199
if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
4200
io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
4201
return pr_manager_execute(s->pr_mgr, qemu_get_current_aio_context(),
4206
acb = (RawPosixAIOData) {
4208
.aio_type = QEMU_AIO_IOCTL,
4209
.aio_fildes = s->fd,
4217
return raw_thread_pool_submit(handle_aiocb_ioctl, &acb);
4221
static coroutine_fn int
4222
hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
4224
BDRVRawState *s = bs->opaque;
4229
raw_account_discard(s, bytes, ret);
4232
return raw_do_pdiscard(bs, offset, bytes, true);
4235
static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
4236
int64_t offset, int64_t bytes, BdrvRequestFlags flags)
4245
return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true);
4248
static BlockDriver bdrv_host_device = {
4249
.format_name = "host_device",
4250
.protocol_name = "host_device",
4251
.instance_size = sizeof(BDRVRawState),
4252
.bdrv_needs_filename = true,
4253
.bdrv_probe_device = hdev_probe_device,
4254
.bdrv_parse_filename = hdev_parse_filename,
4255
.bdrv_open = hdev_open,
4256
.bdrv_close = raw_close,
4257
.bdrv_reopen_prepare = raw_reopen_prepare,
4258
.bdrv_reopen_commit = raw_reopen_commit,
4259
.bdrv_reopen_abort = raw_reopen_abort,
4260
.bdrv_co_create_opts = bdrv_co_create_opts_simple,
4261
.create_opts = &bdrv_create_opts_simple,
4262
.mutable_opts = mutable_opts,
4263
.bdrv_co_invalidate_cache = raw_co_invalidate_cache,
4264
.bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
4266
.bdrv_co_preadv = raw_co_preadv,
4267
.bdrv_co_pwritev = raw_co_pwritev,
4268
.bdrv_co_flush_to_disk = raw_co_flush_to_disk,
4269
.bdrv_co_pdiscard = hdev_co_pdiscard,
4270
.bdrv_co_copy_range_from = raw_co_copy_range_from,
4271
.bdrv_co_copy_range_to = raw_co_copy_range_to,
4272
.bdrv_refresh_limits = raw_refresh_limits,
4274
.bdrv_co_truncate = raw_co_truncate,
4275
.bdrv_co_getlength = raw_co_getlength,
4276
.bdrv_co_get_info = raw_co_get_info,
4277
.bdrv_get_specific_info = raw_get_specific_info,
4278
.bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
4279
.bdrv_get_specific_stats = hdev_get_specific_stats,
4280
.bdrv_check_perm = raw_check_perm,
4281
.bdrv_set_perm = raw_set_perm,
4282
.bdrv_abort_perm_update = raw_abort_perm_update,
4283
.bdrv_probe_blocksizes = hdev_probe_blocksizes,
4284
.bdrv_probe_geometry = hdev_probe_geometry,
4286
/* generic scsi device */
4288
.bdrv_co_ioctl = hdev_co_ioctl,
4292
#if defined(CONFIG_BLKZONED)
4293
/* zone management operations */
4294
.bdrv_co_zone_report = raw_co_zone_report,
4295
.bdrv_co_zone_mgmt = raw_co_zone_mgmt,
4296
.bdrv_co_zone_append = raw_co_zone_append,
4300
#if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
4301
static void cdrom_parse_filename(const char *filename, QDict *options,
4304
bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
4307
static void cdrom_refresh_limits(BlockDriverState *bs, Error **errp)
4309
bs->bl.has_variable_length = true;
4310
raw_refresh_limits(bs, errp);
4315
static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
4318
BDRVRawState *s = bs->opaque;
4322
/* open will not fail even if no CD is inserted, so add O_NONBLOCK */
4323
return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp);
4326
static int cdrom_probe_device(const char *filename)
4332
fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL);
4336
ret = fstat(fd, &st);
4337
if (ret == -1 || !S_ISBLK(st.st_mode)) {
4341
/* Attempt to detect via a CDROM specific ioctl */
4342
ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
4352
static bool coroutine_fn cdrom_co_is_inserted(BlockDriverState *bs)
4354
BDRVRawState *s = bs->opaque;
4357
ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
4358
return ret == CDS_DISC_OK;
4361
static void coroutine_fn cdrom_co_eject(BlockDriverState *bs, bool eject_flag)
4363
BDRVRawState *s = bs->opaque;
4366
if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
4367
perror("CDROMEJECT");
4369
if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
4370
perror("CDROMEJECT");
4374
static void coroutine_fn cdrom_co_lock_medium(BlockDriverState *bs, bool locked)
4376
BDRVRawState *s = bs->opaque;
4378
if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
4380
* Note: an error can happen if the distribution automatically
4383
/* perror("CDROM_LOCKDOOR"); */
4387
static BlockDriver bdrv_host_cdrom = {
4388
.format_name = "host_cdrom",
4389
.protocol_name = "host_cdrom",
4390
.instance_size = sizeof(BDRVRawState),
4391
.bdrv_needs_filename = true,
4392
.bdrv_probe_device = cdrom_probe_device,
4393
.bdrv_parse_filename = cdrom_parse_filename,
4394
.bdrv_open = cdrom_open,
4395
.bdrv_close = raw_close,
4396
.bdrv_reopen_prepare = raw_reopen_prepare,
4397
.bdrv_reopen_commit = raw_reopen_commit,
4398
.bdrv_reopen_abort = raw_reopen_abort,
4399
.bdrv_co_create_opts = bdrv_co_create_opts_simple,
4400
.create_opts = &bdrv_create_opts_simple,
4401
.mutable_opts = mutable_opts,
4402
.bdrv_co_invalidate_cache = raw_co_invalidate_cache,
4404
.bdrv_co_preadv = raw_co_preadv,
4405
.bdrv_co_pwritev = raw_co_pwritev,
4406
.bdrv_co_flush_to_disk = raw_co_flush_to_disk,
4407
.bdrv_refresh_limits = cdrom_refresh_limits,
4409
.bdrv_co_truncate = raw_co_truncate,
4410
.bdrv_co_getlength = raw_co_getlength,
4411
.bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
4413
/* removable device support */
4414
.bdrv_co_is_inserted = cdrom_co_is_inserted,
4415
.bdrv_co_eject = cdrom_co_eject,
4416
.bdrv_co_lock_medium = cdrom_co_lock_medium,
4418
/* generic scsi device */
4419
.bdrv_co_ioctl = hdev_co_ioctl,
4421
#endif /* __linux__ */
4423
#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
4424
static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
4427
BDRVRawState *s = bs->opaque;
4432
ret = raw_open_common(bs, options, flags, 0, true, errp);
4437
/* make sure the door isn't locked at this time */
4438
ioctl(s->fd, CDIOCALLOW);
4442
static int cdrom_probe_device(const char *filename)
4444
if (strstart(filename, "/dev/cd", NULL) ||
4445
strstart(filename, "/dev/acd", NULL))
4450
static int cdrom_reopen(BlockDriverState *bs)
4452
BDRVRawState *s = bs->opaque;
4456
* Force reread of possibly changed/newly loaded disc,
4457
* FreeBSD seems to not notice sometimes...
4461
fd = qemu_open(bs->filename, s->open_flags, NULL);
4468
/* make sure the door isn't locked at this time */
4469
ioctl(s->fd, CDIOCALLOW);
4473
static bool coroutine_fn cdrom_co_is_inserted(BlockDriverState *bs)
4475
return raw_getlength(bs) > 0;
4478
static void coroutine_fn cdrom_co_eject(BlockDriverState *bs, bool eject_flag)
4480
BDRVRawState *s = bs->opaque;
4485
(void) ioctl(s->fd, CDIOCALLOW);
4488
if (ioctl(s->fd, CDIOCEJECT) < 0)
4489
perror("CDIOCEJECT");
4491
if (ioctl(s->fd, CDIOCCLOSE) < 0)
4492
perror("CDIOCCLOSE");
4498
static void coroutine_fn cdrom_co_lock_medium(BlockDriverState *bs, bool locked)
4500
BDRVRawState *s = bs->opaque;
4504
if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
4506
* Note: an error can happen if the distribution automatically
4509
/* perror("CDROM_LOCKDOOR"); */
4513
static BlockDriver bdrv_host_cdrom = {
4514
.format_name = "host_cdrom",
4515
.protocol_name = "host_cdrom",
4516
.instance_size = sizeof(BDRVRawState),
4517
.bdrv_needs_filename = true,
4518
.bdrv_probe_device = cdrom_probe_device,
4519
.bdrv_parse_filename = cdrom_parse_filename,
4520
.bdrv_open = cdrom_open,
4521
.bdrv_close = raw_close,
4522
.bdrv_reopen_prepare = raw_reopen_prepare,
4523
.bdrv_reopen_commit = raw_reopen_commit,
4524
.bdrv_reopen_abort = raw_reopen_abort,
4525
.bdrv_co_create_opts = bdrv_co_create_opts_simple,
4526
.create_opts = &bdrv_create_opts_simple,
4527
.mutable_opts = mutable_opts,
4529
.bdrv_co_preadv = raw_co_preadv,
4530
.bdrv_co_pwritev = raw_co_pwritev,
4531
.bdrv_co_flush_to_disk = raw_co_flush_to_disk,
4532
.bdrv_refresh_limits = cdrom_refresh_limits,
4534
.bdrv_co_truncate = raw_co_truncate,
4535
.bdrv_co_getlength = raw_co_getlength,
4536
.bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
4538
/* removable device support */
4539
.bdrv_co_is_inserted = cdrom_co_is_inserted,
4540
.bdrv_co_eject = cdrom_co_eject,
4541
.bdrv_co_lock_medium = cdrom_co_lock_medium,
4543
#endif /* __FreeBSD__ */
4545
#endif /* HAVE_HOST_BLOCK_DEVICE */
4547
static void bdrv_file_init(void)
4550
* Register all the drivers. Note that order is important, the driver
4551
* registered last will get probed first.
4553
bdrv_register(&bdrv_file);
4554
#if defined(HAVE_HOST_BLOCK_DEVICE)
4555
bdrv_register(&bdrv_host_device);
4557
bdrv_register(&bdrv_host_cdrom);
4559
#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
4560
bdrv_register(&bdrv_host_cdrom);
4562
#endif /* HAVE_HOST_BLOCK_DEVICE */
4565
block_init(bdrv_file_init);