*
* Implements a one-block write-through cache.
*
- * Includes support for Windows NT support under Cygwin.
+ * Includes support for Windows NT support under Cygwin.
*
* Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
* 2002 by Theodore Ts'o.
*
* %Begin-Header%
- * This file may be redistributed under the terms of the GNU Public
- * License.
+ * This file may be redistributed under the terms of the GNU Library
+ * General Public License, version 2.
* %End-Header%
*/
#define _LARGEFILE_SOURCE
#define _LARGEFILE64_SOURCE
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
#include <stdio.h>
#include <string.h>
#ifdef __linux__
#include <sys/utsname.h>
#endif
+#ifdef HAVE_SYS_IOCTL_H
+#include <sys/ioctl.h>
+#endif
+#ifdef HAVE_SYS_MOUNT_H
+#include <sys/mount.h>
+#endif
#if HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#include <sys/resource.h>
#endif
+#if defined(__linux__) && defined(_IO) && !defined(BLKROGET)
+#define BLKROGET _IO(0x12, 94) /* Get read-only status (0 = read_write). */
+#endif
+
+#if defined(__linux__) && defined(_IO) && !defined(BLKSSZGET)
+#define BLKSSZGET _IO(0x12,104)/* get block device sector size */
+#endif
+
+#undef ALIGN_DEBUG
+
#include "ext2_fs.h"
#include "ext2fs.h"
int magic;
int dev;
int flags;
+ int align;
int access_time;
ext2_loff_t offset;
struct unix_cache cache[CACHE_SIZE];
+ void *bounce;
+ struct struct_io_stats io_stats;
};
+#define IS_ALIGNED(n, align) ((((unsigned long) n) & \
+ ((unsigned long) ((align)-1))) == 0)
+
static errcode_t unix_open(const char *name, int flags, io_channel *channel);
static errcode_t unix_close(io_channel channel);
static errcode_t unix_set_blksize(io_channel channel, int blksize);
static errcode_t unix_flush(io_channel channel);
static errcode_t unix_write_byte(io_channel channel, unsigned long offset,
int size, const void *data);
-static errcode_t unix_set_option(io_channel channel, const char *option,
+static errcode_t unix_set_option(io_channel channel, const char *option,
const char *arg);
-
+static errcode_t unix_get_stats(io_channel channel, io_stats *stats)
+;
static void reuse_cache(io_channel channel, struct unix_private_data *data,
- struct unix_cache *cache, unsigned long block);
-
-/* __FreeBSD_kernel__ is defined by GNU/kFreeBSD - the FreeBSD kernel
- * does not know buffered block devices - everything is raw. */
-#if defined(__CYGWIN__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
-#define NEED_BOUNCE_BUFFER
-#else
-#undef NEED_BOUNCE_BUFFER
-#endif
+ struct unix_cache *cache, unsigned long long block);
+static errcode_t unix_read_blk64(io_channel channel, unsigned long long block,
+ int count, void *data);
+static errcode_t unix_write_blk64(io_channel channel, unsigned long long block,
+ int count, const void *data);
+static errcode_t unix_discard(io_channel channel, unsigned long long block,
+ unsigned long long count);
static struct struct_io_manager struct_unix_manager = {
EXT2_ET_MAGIC_IO_MANAGER,
unix_read_blk,
unix_write_blk,
unix_flush,
-#ifdef NEED_BOUNCE_BUFFER
- 0,
-#else
unix_write_byte,
-#endif
- unix_set_option
+ unix_set_option,
+ unix_get_stats,
+ unix_read_blk64,
+ unix_write_blk64,
+ unix_discard,
};
io_manager unix_io_manager = &struct_unix_manager;
+static errcode_t unix_get_stats(io_channel channel, io_stats *stats)
+{
+ errcode_t retval = 0;
+
+ struct unix_private_data *data;
+
+ EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+ data = (struct unix_private_data *) channel->private_data;
+ EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+ if (stats)
+ *stats = &data->io_stats;
+
+ return retval;
+}
+
/*
* Here are the raw I/O functions
*/
-#ifndef NEED_BOUNCE_BUFFER
static errcode_t raw_read_blk(io_channel channel,
struct unix_private_data *data,
- unsigned long block,
+ unsigned long long block,
int count, void *buf)
{
errcode_t retval;
int actual = 0;
size = (count < 0) ? -count : count * channel->block_size;
+ data->io_stats.bytes_read += size;
location = ((ext2_loff_t) block * channel->block_size) + data->offset;
if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) {
retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
goto error_out;
}
- actual = read(data->dev, buf, size);
- if (actual != size) {
- if (actual < 0)
- actual = 0;
- retval = EXT2_ET_SHORT_READ;
- goto error_out;
+ if ((data->align == 0) ||
+ ((IS_ALIGNED(buf, data->align)) && IS_ALIGNED(size, data->align))) {
+ actual = read(data->dev, buf, size);
+ if (actual != size) {
+ short_read:
+ if (actual < 0)
+ actual = 0;
+ retval = EXT2_ET_SHORT_READ;
+ goto error_out;
+ }
+ return 0;
}
- return 0;
-
-error_out:
- memset((char *) buf+actual, 0, size-actual);
- if (channel->read_error)
- retval = (channel->read_error)(channel, block, count, buf,
- size, actual, retval);
- return retval;
-}
-#else /* NEED_BOUNCE_BUFFER */
-/*
- * Windows and FreeBSD block devices only allow sector alignment IO in offset and size
- */
-static errcode_t raw_read_blk(io_channel channel,
- struct unix_private_data *data,
- unsigned long block,
- int count, void *buf)
-{
- errcode_t retval;
- size_t size, alignsize, fragment;
- ext2_loff_t location;
- int total = 0, actual;
-#define BLOCKALIGN 512
- char sector[BLOCKALIGN];
- size = (count < 0) ? -count : count * channel->block_size;
- location = ((ext2_loff_t) block * channel->block_size) + data->offset;
-#ifdef DEBUG
- printf("count=%d, size=%d, block=%d, blk_size=%d, location=%lx\n",
- count, size, block, channel->block_size, location);
+#ifdef ALIGN_DEBUG
+ printf("raw_read_blk: O_DIRECT fallback: %p %lu\n", buf,
+ (unsigned long) size);
#endif
- if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) {
- retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
- goto error_out;
- }
- fragment = size % BLOCKALIGN;
- alignsize = size - fragment;
- if (alignsize) {
- actual = read(data->dev, buf, alignsize);
- if (actual != alignsize)
- goto short_read;
- }
- if (fragment) {
- actual = read(data->dev, sector, BLOCKALIGN);
- if (actual != BLOCKALIGN)
+
+ /*
+ * The buffer or size which we're trying to read isn't aligned
+ * to the O_DIRECT rules, so we need to do this the hard way...
+ */
+ while (size > 0) {
+ actual = read(data->dev, data->bounce, channel->block_size);
+ if (actual != channel->block_size)
goto short_read;
- memcpy(buf+alignsize, sector, fragment);
+ actual = size;
+ if (size > channel->block_size)
+ actual = channel->block_size;
+ memcpy(buf, data->bounce, actual);
+ size -= actual;
+ buf += actual;
}
return 0;
-short_read:
- if (actual>0)
- total += actual;
- retval = EXT2_ET_SHORT_READ;
-
error_out:
- memset((char *) buf+total, 0, size-actual);
+ memset((char *) buf+actual, 0, size-actual);
if (channel->read_error)
retval = (channel->read_error)(channel, block, count, buf,
size, actual, retval);
return retval;
}
-#endif
static errcode_t raw_write_blk(io_channel channel,
struct unix_private_data *data,
- unsigned long block,
+ unsigned long long block,
int count, const void *buf)
{
ssize_t size;
else
size = count * channel->block_size;
}
+ data->io_stats.bytes_written += size;
location = ((ext2_loff_t) block * channel->block_size) + data->offset;
if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) {
retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
goto error_out;
}
-
- actual = write(data->dev, buf, size);
- if (actual != size) {
- retval = EXT2_ET_SHORT_WRITE;
- goto error_out;
+
+ if ((data->align == 0) ||
+ ((IS_ALIGNED(buf, data->align)) && IS_ALIGNED(size, data->align))) {
+ actual = write(data->dev, buf, size);
+ if (actual != size) {
+ short_write:
+ retval = EXT2_ET_SHORT_WRITE;
+ goto error_out;
+ }
+ return 0;
+ }
+
+#ifdef ALIGN_DEBUG
+ printf("raw_write_blk: O_DIRECT fallback: %p %lu\n", buf,
+ (unsigned long) size);
+#endif
+ /*
+ * The buffer or size which we're trying to write isn't aligned
+ * to the O_DIRECT rules, so we need to do this the hard way...
+ */
+ while (size > 0) {
+ if (size < channel->block_size) {
+ actual = read(data->dev, data->bounce,
+ channel->block_size);
+ if (actual != channel->block_size) {
+ retval = EXT2_ET_SHORT_READ;
+ goto error_out;
+ }
+ }
+ actual = size;
+ if (size > channel->block_size)
+ actual = channel->block_size;
+ memcpy(data->bounce, buf, actual);
+ actual = write(data->dev, data->bounce, channel->block_size);
+ if (actual != channel->block_size)
+ goto short_write;
+ size -= actual;
+ buf += actual;
}
return 0;
-
+
error_out:
if (channel->write_error)
retval = (channel->write_error)(channel, block, count, buf,
errcode_t retval;
struct unix_cache *cache;
int i;
-
+
data->access_time = 0;
for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
cache->block = 0;
cache->access_time = 0;
cache->dirty = 0;
cache->in_use = 0;
- if ((retval = ext2fs_get_mem(channel->block_size,
- &cache->buf)))
+ if (cache->buf)
+ ext2fs_free_mem(&cache->buf);
+ retval = ext2fs_get_memalign(channel->block_size,
+ data->align, &cache->buf);
+ if (retval)
return retval;
}
- return 0;
+ if (data->align) {
+ if (data->bounce)
+ ext2fs_free_mem(&data->bounce);
+ retval = ext2fs_get_memalign(channel->block_size, data->align,
+ &data->bounce);
+ }
+ return retval;
}
/* Free the cache buffers */
{
struct unix_cache *cache;
int i;
-
+
data->access_time = 0;
for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
cache->block = 0;
cache->in_use = 0;
if (cache->buf)
ext2fs_free_mem(&cache->buf);
- cache->buf = 0;
}
+ if (data->bounce)
+ ext2fs_free_mem(&data->bounce);
}
#ifndef NO_IO_CACHE
* entry to that should be reused.
*/
static struct unix_cache *find_cached_block(struct unix_private_data *data,
- unsigned long block,
+ unsigned long long block,
struct unix_cache **eldest)
{
struct unix_cache *cache, *unused_cache, *oldest_cache;
int i;
-
+
unused_cache = oldest_cache = 0;
for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
if (!cache->in_use) {
* Reuse a particular cache entry for another block.
*/
static void reuse_cache(io_channel channel, struct unix_private_data *data,
- struct unix_cache *cache, unsigned long block)
+ struct unix_cache *cache, unsigned long long block)
{
if (cache->dirty && cache->in_use)
raw_write_blk(channel, data, cache->block, 1, cache->buf);
struct unix_cache *cache;
errcode_t retval, retval2;
int i;
-
+
retval2 = 0;
for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
if (!cache->in_use)
continue;
-
+
if (invalidate)
cache->in_use = 0;
-
+
if (!cache->dirty)
continue;
-
+
retval = raw_write_blk(channel, data,
cache->block, 1, cache->buf);
if (retval)
}
#endif /* NO_IO_CACHE */
+#ifdef __linux__
+#ifndef BLKDISCARDZEROES
+#define BLKDISCARDZEROES _IO(0x12,124)
+#endif
+#endif
+
static errcode_t unix_open(const char *name, int flags, io_channel *channel)
{
io_channel io = NULL;
struct unix_private_data *data = NULL;
errcode_t retval;
- int open_flags;
+ int open_flags, zeroes = 0;
struct stat st;
#ifdef __linux__
struct utsname ut;
memset(data, 0, sizeof(struct unix_private_data));
data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL;
-
- if ((retval = alloc_cache(io, data)))
- goto cleanup;
+ data->io_stats.num_fields = 2;
open_flags = (flags & IO_FLAG_RW) ? O_RDWR : O_RDONLY;
if (flags & IO_FLAG_EXCLUSIVE)
open_flags |= O_EXCL;
+#ifdef O_DIRECT
+ if (flags & IO_FLAG_DIRECT_IO)
+ open_flags |= O_DIRECT;
+#endif
+ data->flags = flags;
+
#ifdef HAVE_OPEN64
data->dev = open64(io->name, open_flags);
#else
goto cleanup;
}
+#ifdef BLKSSZGET
+ if (flags & IO_FLAG_DIRECT_IO) {
+ if (ioctl(data->dev, BLKSSZGET, &data->align) != 0)
+ data->align = io->block_size;
+ }
+#endif
+
+#ifdef BLKDISCARDZEROES
+ ioctl(data->dev, BLKDISCARDZEROES, &zeroes);
+ if (zeroes)
+ io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES;
+#endif
+
+#if defined(__CYGWIN__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
+ /*
+ * Some operating systems require that the buffers be aligned,
+ * regardless of O_DIRECT
+ */
+ data->align = 512;
+#endif
+
+
+ if ((retval = alloc_cache(io, data)))
+ goto cleanup;
+
+#ifdef BLKROGET
+ if (flags & IO_FLAG_RW) {
+ int error;
+ int readonly = 0;
+
+ /* Is the block device actually writable? */
+ error = ioctl(data->dev, BLKROGET, &readonly);
+ if (!error && readonly) {
+ close(data->dev);
+ retval = EPERM;
+ goto cleanup;
+ }
+ }
+#endif
+
#ifdef __linux__
#undef RLIM_INFINITY
#if (defined(__alpha__) || ((defined(__sparc__) || defined(__mips__)) && (SIZEOF_LONG == 4)))
* block devices are wrongly getting hit by the filesize
* limit. This workaround isn't perfect, since it won't work
* if glibc wasn't built against 2.2 header files. (Sigh.)
- *
+ *
*/
if ((flags & IO_FLAG_RW) &&
(uname(&ut) == 0) &&
(fstat(data->dev, &st) == 0) &&
(S_ISBLK(st.st_mode))) {
struct rlimit rlim;
-
+
rlim.rlim_cur = rlim.rlim_max = (unsigned long) RLIM_INFINITY;
setrlimit(RLIMIT_FSIZE, &rlim);
getrlimit(RLIMIT_FSIZE, &rlim);
if ((retval = flush_cached_blocks(channel, data, 0)))
return retval;
#endif
-
+
channel->block_size = blksize;
free_cache(data);
if ((retval = alloc_cache(channel, data)))
}
-static errcode_t unix_read_blk(io_channel channel, unsigned long block,
+static errcode_t unix_read_blk64(io_channel channel, unsigned long long block,
int count, void *buf)
{
struct unix_private_data *data;
/* If it's in the cache, use it! */
if ((cache = find_cached_block(data, block, &reuse[0]))) {
#ifdef DEBUG
- printf("Using cached block %d\n", block);
+ printf("Using cached block %lu\n", block);
#endif
memcpy(cp, cache->buf, channel->block_size);
count--;
cp += channel->block_size;
continue;
}
+ if (count == 1) {
+ /*
+ * Special case where we read directly into the
+ * cache buffer; important in the O_DIRECT case
+ */
+ cache = reuse[0];
+ reuse_cache(channel, data, cache, block);
+ if ((retval = raw_read_blk(channel, data, block, 1,
+ cache->buf))) {
+ cache->in_use = 0;
+ return retval;
+ }
+ memcpy(cp, cache->buf, channel->block_size);
+ return 0;
+ }
+
/*
* Find the number of uncached blocks so we can do a
* single read request
if (find_cached_block(data, block+i, &reuse[i]))
break;
#ifdef DEBUG
- printf("Reading %d blocks starting at %d\n", i, block);
+ printf("Reading %d blocks starting at %lu\n", i, block);
#endif
if ((retval = raw_read_blk(channel, data, block, i, cp)))
return retval;
-
+
/* Save the results in the cache */
for (j=0; j < i; j++) {
count--;
#endif /* NO_IO_CACHE */
}
-static errcode_t unix_write_blk(io_channel channel, unsigned long block,
+static errcode_t unix_read_blk(io_channel channel, unsigned long block,
+ int count, void *buf)
+{
+ return unix_read_blk64(channel, block, count, buf);
+}
+
+static errcode_t unix_write_blk64(io_channel channel, unsigned long long block,
int count, const void *buf)
{
struct unix_private_data *data;
#ifdef NO_IO_CACHE
return raw_write_blk(channel, data, block, count, buf);
-#else
+#else
/*
* If we're doing an odd-sized write or a very large write,
* flush out the cache completely and then do a direct write.
writethrough = channel->flags & CHANNEL_FLAGS_WRITETHROUGH;
if (writethrough)
retval = raw_write_blk(channel, data, block, count, buf);
-
+
cp = buf;
while (count > 0) {
cache = find_cached_block(data, block, &reuse);
#endif /* NO_IO_CACHE */
}
+static errcode_t unix_write_blk(io_channel channel, unsigned long block,
+ int count, const void *buf)
+{
+ return unix_write_blk64(channel, block, count, buf);
+}
+
static errcode_t unix_write_byte(io_channel channel, unsigned long offset,
int size, const void *buf)
{
data = (struct unix_private_data *) channel->private_data;
EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+ if (data->align != 0) {
+#ifdef ALIGN_DEBUG
+ printf("unix_write_byte: O_DIRECT fallback\n");
+#endif
+ return EXT2_ET_UNIMPLEMENTED;
+ }
+
#ifndef NO_IO_CACHE
/*
* Flush out the cache completely
if (lseek(data->dev, offset + data->offset, SEEK_SET) < 0)
return errno;
-
+
actual = write(data->dev, buf, size);
if (actual != size)
return EXT2_ET_SHORT_WRITE;
}
/*
- * Flush data buffers to disk.
+ * Flush data buffers to disk.
*/
static errcode_t unix_flush(io_channel channel)
{
struct unix_private_data *data;
errcode_t retval = 0;
-
+
EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
data = (struct unix_private_data *) channel->private_data;
EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
return retval;
}
-static errcode_t unix_set_option(io_channel channel, const char *option,
+static errcode_t unix_set_option(io_channel channel, const char *option,
const char *arg)
{
struct unix_private_data *data;
- unsigned long tmp;
+ unsigned long long tmp;
char *end;
EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
if (!arg)
return EXT2_ET_INVALID_ARGUMENT;
- tmp = strtoul(arg, &end, 0);
+ tmp = strtoull(arg, &end, 0);
if (*end)
return EXT2_ET_INVALID_ARGUMENT;
data->offset = tmp;
+ if (data->offset < 0)
+ return EXT2_ET_INVALID_ARGUMENT;
return 0;
}
return EXT2_ET_INVALID_ARGUMENT;
}
+
+#if defined(__linux__) && !defined(BLKDISCARD)
+#define BLKDISCARD _IO(0x12,119)
+#endif
+
+static errcode_t unix_discard(io_channel channel, unsigned long long block,
+ unsigned long long count)
+{
+#ifdef BLKDISCARD
+ struct unix_private_data *data;
+ __uint64_t range[2];
+ int ret;
+
+ EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+ data = (struct unix_private_data *) channel->private_data;
+ EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+ range[0] = (__uint64_t)(block) * channel->block_size;
+ range[1] = (__uint64_t)(count) * channel->block_size;
+
+ ret = ioctl(data->dev, BLKDISCARD, &range);
+ if (ret < 0)
+ return errno;
+ return 0;
+#else
+ return EXT2_ET_UNIMPLEMENTED;
+#endif
+}