raw: Probe required direct I/O alignment

Add a bs->request_alignment field that contains the required
offset/length alignment for I/O requests and fill it in the raw block
drivers. Use ioctls if possible, else see what alignment it takes for
O_DIRECT to succeed.

While at it, also expose the memory alignment requirements, which may be
(and in practice are) different from the disk alignment requirements.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
This commit is contained in:
Paolo Bonzini 2011-11-29 12:42:20 +01:00 committed by Kevin Wolf
parent 1b7fd72955
commit c25f53b06e
4 changed files with 132 additions and 17 deletions

View File

@ -852,6 +852,7 @@ static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
bs->open_flags = flags;
bs->guest_block_size = 512;
bs->request_alignment = 512;
bs->zero_beyond_eof = true;
open_flags = bdrv_open_flags(bs, flags);
bs->read_only = !(open_flags & BDRV_O_RDWR);
@ -920,6 +921,8 @@ static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
}
bdrv_refresh_limits(bs);
assert(bdrv_opt_mem_align(bs) != 0);
assert(bs->request_alignment != 0);
#ifndef _WIN32
if (bs->is_temporary) {

View File

@ -127,6 +127,8 @@ typedef struct BDRVRawState {
int fd;
int type;
int open_flags;
size_t buf_align;
#if defined(__linux__)
/* linux floppy specific */
int64_t fd_open_time;
@ -213,6 +215,76 @@ static int raw_normalize_devicepath(const char **filename)
}
#endif
static void raw_probe_alignment(BlockDriverState *bs)
{
BDRVRawState *s = bs->opaque;
char *buf;
unsigned int sector_size;
/* For /dev/sg devices the alignment is not really used.
With buffered I/O, we don't have any restrictions. */
if (bs->sg || !(s->open_flags & O_DIRECT)) {
bs->request_alignment = 1;
s->buf_align = 1;
return;
}
/* Try a few ioctls to get the right size */
bs->request_alignment = 0;
s->buf_align = 0;
#ifdef BLKSSZGET
if (ioctl(s->fd, BLKSSZGET, &sector_size) >= 0) {
bs->request_alignment = sector_size;
}
#endif
#ifdef DKIOCGETBLOCKSIZE
if (ioctl(s->fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
bs->request_alignment = sector_size;
}
#endif
#ifdef DIOCGSECTORSIZE
if (ioctl(s->fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
bs->request_alignment = sector_size;
}
#endif
#ifdef CONFIG_XFS
if (s->is_xfs) {
struct dioattr da;
if (xfsctl(NULL, s->fd, XFS_IOC_DIOINFO, &da) >= 0) {
bs->request_alignment = da.d_miniosz;
/* The kernel returns wrong information for d_mem */
/* s->buf_align = da.d_mem; */
}
}
#endif
/* If we could not get the sizes so far, we can only guess them */
if (!s->buf_align) {
size_t align;
buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE);
for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
if (pread(s->fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) {
s->buf_align = align;
break;
}
}
qemu_vfree(buf);
}
if (!bs->request_alignment) {
size_t align;
buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE);
for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
if (pread(s->fd, buf, align, 0) >= 0) {
bs->request_alignment = align;
break;
}
}
qemu_vfree(buf);
}
}
static void raw_parse_flags(int bdrv_flags, int *open_flags)
{
assert(open_flags != NULL);
@ -463,7 +535,6 @@ static int raw_reopen_prepare(BDRVReopenState *state,
return ret;
}
static void raw_reopen_commit(BDRVReopenState *state)
{
BDRVRawReopenState *raw_s = state->opaque;
@ -499,23 +570,15 @@ static void raw_reopen_abort(BDRVReopenState *state)
state->opaque = NULL;
}
static int raw_refresh_limits(BlockDriverState *bs)
{
BDRVRawState *s = bs->opaque;
/* XXX: use host sector size if necessary with:
#ifdef DIOCGSECTORSIZE
{
unsigned int sectorsize = 512;
if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
sectorsize > bufsize)
bufsize = sectorsize;
}
#endif
#ifdef CONFIG_COCOA
uint32_t blockSize = 512;
if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
bufsize = blockSize;
}
#endif
*/
raw_probe_alignment(bs);
bs->bl.opt_mem_alignment = s->buf_align;
return 0;
}
static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
{
@ -1363,6 +1426,7 @@ static BlockDriver bdrv_file = {
.bdrv_aio_writev = raw_aio_writev,
.bdrv_aio_flush = raw_aio_flush,
.bdrv_aio_discard = raw_aio_discard,
.bdrv_refresh_limits = raw_refresh_limits,
.bdrv_truncate = raw_truncate,
.bdrv_getlength = raw_getlength,
@ -1740,6 +1804,7 @@ static BlockDriver bdrv_host_device = {
.bdrv_aio_writev = raw_aio_writev,
.bdrv_aio_flush = raw_aio_flush,
.bdrv_aio_discard = hdev_aio_discard,
.bdrv_refresh_limits = raw_refresh_limits,
.bdrv_truncate = raw_truncate,
.bdrv_getlength = raw_getlength,
@ -1871,6 +1936,7 @@ static BlockDriver bdrv_host_floppy = {
.bdrv_aio_readv = raw_aio_readv,
.bdrv_aio_writev = raw_aio_writev,
.bdrv_aio_flush = raw_aio_flush,
.bdrv_refresh_limits = raw_refresh_limits,
.bdrv_truncate = raw_truncate,
.bdrv_getlength = raw_getlength,
@ -1981,6 +2047,7 @@ static BlockDriver bdrv_host_cdrom = {
.bdrv_aio_readv = raw_aio_readv,
.bdrv_aio_writev = raw_aio_writev,
.bdrv_aio_flush = raw_aio_flush,
.bdrv_refresh_limits = raw_refresh_limits,
.bdrv_truncate = raw_truncate,
.bdrv_getlength = raw_getlength,
@ -2110,6 +2177,7 @@ static BlockDriver bdrv_host_cdrom = {
.bdrv_aio_readv = raw_aio_readv,
.bdrv_aio_writev = raw_aio_writev,
.bdrv_aio_flush = raw_aio_flush,
.bdrv_refresh_limits = raw_refresh_limits,
.bdrv_truncate = raw_truncate,
.bdrv_getlength = raw_getlength,

View File

@ -202,6 +202,35 @@ static int set_sparse(int fd)
NULL, 0, NULL, 0, &returned, NULL);
}
static void raw_probe_alignment(BlockDriverState *bs)
{
BDRVRawState *s = bs->opaque;
DWORD sectorsPerCluster, freeClusters, totalClusters, count;
DISK_GEOMETRY_EX dg;
BOOL status;
if (s->type == FTYPE_CD) {
bs->request_alignment = 2048;
return;
}
if (s->type == FTYPE_HARDDISK) {
status = DeviceIoControl(s->hfile, IOCTL_DISK_GET_DRIVE_GEOMETRY_EX,
NULL, 0, &dg, sizeof(dg), &count, NULL);
if (status != 0) {
bs->request_alignment = dg.Geometry.BytesPerSector;
return;
}
/* try GetDiskFreeSpace too */
}
if (s->drive_path[0]) {
GetDiskFreeSpace(s->drive_path, &sectorsPerCluster,
&dg.Geometry.BytesPerSector,
&freeClusters, &totalClusters);
bs->request_alignment = dg.Geometry.BytesPerSector;
}
}
static void raw_parse_flags(int flags, int *access_flags, DWORD *overlapped)
{
assert(access_flags != NULL);
@ -269,6 +298,17 @@ static int raw_open(BlockDriverState *bs, QDict *options, int flags,
}
}
if (filename[0] && filename[1] == ':') {
snprintf(s->drive_path, sizeof(s->drive_path), "%c:\\", filename[0]);
} else if (filename[0] == '\\' && filename[1] == '\\') {
s->drive_path[0] = 0;
} else {
/* Relative path. */
char buf[MAX_PATH];
GetCurrentDirectory(MAX_PATH, buf);
snprintf(s->drive_path, sizeof(s->drive_path), "%c:\\", buf[0]);
}
s->hfile = CreateFile(filename, access_flags,
FILE_SHARE_READ, NULL,
OPEN_EXISTING, overlapped, NULL);
@ -293,6 +333,7 @@ static int raw_open(BlockDriverState *bs, QDict *options, int flags,
s->aio = aio;
}
raw_probe_alignment(bs);
ret = 0;
fail:
qemu_opts_del(opts);

View File

@ -325,6 +325,9 @@ struct BlockDriverState {
/* Whether produces zeros when read beyond eof */
bool zero_beyond_eof;
/* Alignment requirement for offset/length of I/O requests */
unsigned int request_alignment;
/* the block size for which the guest device expects atomicity */
int guest_block_size;