Hi,
Recently there was discussion about an "FIle Extent MAP"(FIEMAP) ioctl for efficiently mapping the extents and holes of a file. This will be many times more efficient than FIBMAP by cutting down the number of ioctls.
This patch implements this ioctl. Ted, in this patch I have used ioctl nunber 10 for FIEMAP. We need to make sure that the ioctl number in our releases and upstream is same. So can you please advise if we can register ioctl no 10 for FIEMAP or if I need to change it?
Let me just mention the the structures and flags used in FIEMAP for people to have a quick review:
struct fiemap_extent {
__u64 fe_offset; /* offset in bytes for the start of the extent */
__u64 fe_length; /* length in bytes for the extent */
__u32 fe_flags; /* returned FIEMAP_EXTENT_* flags for the extent */
__u32 fe_lun; /* logical device number for extent (starting at 0)*/
};
/*
* fiemap is not ext4-specific and should be moved into fs.h eventually.
*/
struct fiemap {
__u64 fm_start; /* logical starting byte offset (in/out) */
__u64 fm_length; /* logical length of map (in/out) */
__u32 fm_flags; /* FIEMAP_FLAG_* flags for request (in/out) */
__u32 fm_extent_count; /* number of extents in fm_extents (in/out) */
__u64 fm_unused;
struct fiemap_extent fm_extents[0];
};
#define FIEMAP_FLAG_SYNC 0x00000001 /* sync file data before map */
#define FIEMAP_FLAG_HSM_READ 0x00000002 /* get data from HSM before map */
#define FIEMAP_FLAG_NUM_EXTENTS 0x00000004 /* return only number of extents */
#define FIEMAP_FLAG_INCOMPAT 0xff000000 /* error for unknown flags in here */
#define FIEMAP_EXTENT_HOLE 0x00000001 /* has no data or space allocation */
#define FIEMAP_EXTENT_UNWRITTEN 0x00000002 /* space allocated, but no data */
#define FIEMAP_EXTENT_UNMAPPED 0x00000004 /* has data but no space allocation*/
#define FIEMAP_EXTENT_ERROR 0x00000008 /* mapping error, errno in fe_start*/
#define FIEMAP_EXTENT_NO_DIRECT 0x00000010 /* cannot access data directly */
#define FIEMAP_EXTENT_LAST 0x00000020 /* last extent in the file */
#define FIEMAP_EXTENT_DELALLOC 0x00000040 /* has data but not yet written,
* must have EXTENT_UNKNOWN set */
#define FIEMAP_EXTENT_SECONDARY 0x00000080 /* data (also) in secondary storage,
* not in primary if EXTENT_UNKNOWN*/
#define FIEMAP_EXTENT_EOF 0x00000100 /* if fm_start+fm_len is beyond EOF*/
--
checkpatch.pl has been run on this patch and did not report any errors.
Signed-off-by: Andreas Dilger <adilger@xxxxxxxxxxxxx>
Signed-off-by: Kalpak Shah <kalpak@xxxxxxxxxxxxx>
Thanks,
Kalpak.
Index: linux-2.6.22.1/fs/ext4/ioctl.c
===================================================================
--- linux-2.6.22.1.orig/fs/ext4/ioctl.c
+++ linux-2.6.22.1/fs/ext4/ioctl.c
@@ -16,6 +16,7 @@
#include <linux/compat.h>
#include <linux/smp_lock.h>
#include <asm/uaccess.h>
+#include "fiemap.h"
int ext4_ioctl (struct inode * inode, struct file * filp, unsigned int cmd,
unsigned long arg)
@@ -248,6 +249,9 @@ flags_err:
return err;
}
+ case EXT4_IOC_FIEMAP: {
+ return ext4_fiemap(inode, filp, cmd, arg);
+ }
default:
return -ENOTTY;
Index: linux-2.6.22.1/include/linux/ext4_fs.h
===================================================================
--- linux-2.6.22.1.orig/include/linux/ext4_fs.h
+++ linux-2.6.22.1/include/linux/ext4_fs.h
@@ -240,15 +240,16 @@ struct ext4_new_group_data {
#define EXT4_IOC_SETFLAGS FS_IOC_SETFLAGS
#define EXT4_IOC_GETVERSION _IOR('f', 3, long)
#define EXT4_IOC_SETVERSION _IOW('f', 4, long)
+#define EXT4_IOC_GETRSVSZ _IOR('f', 5, long)
+#define EXT4_IOC_SETRSVSZ _IOW('f', 6, long)
#define EXT4_IOC_GROUP_EXTEND _IOW('f', 7, unsigned long)
#define EXT4_IOC_GROUP_ADD _IOW('f', 8,struct ext4_new_group_input)
+#define EXT4_IOC_FIEMAP _IOWR('f', 10, struct fiemap)
#define EXT4_IOC_GETVERSION_OLD FS_IOC_GETVERSION
#define EXT4_IOC_SETVERSION_OLD FS_IOC_SETVERSION
#ifdef CONFIG_JBD2_DEBUG
#define EXT4_IOC_WAIT_FOR_READONLY _IOR('f', 99, long)
#endif
-#define EXT4_IOC_GETRSVSZ _IOR('f', 5, long)
-#define EXT4_IOC_SETRSVSZ _IOW('f', 6, long)
/*
* ioctl commands in 32 bit emulation
@@ -266,7 +267,6 @@ struct ext4_new_group_data {
#define EXT4_IOC32_GETVERSION_OLD FS_IOC32_GETVERSION
#define EXT4_IOC32_SETVERSION_OLD FS_IOC32_SETVERSION
-
/*
* Mount options
*/
@@ -1094,6 +1094,8 @@ ext4_get_blocks_wrap(handle_t *handle, s
return ext4_get_blocks_handle(handle, inode, block, max_blocks, bh,
create, extend_disksize);
}
+extern int ext4_fiemap(struct inode *, struct file *, unsigned int,
+ unsigned long);
#endif /* __KERNEL__ */
Index: linux-2.6.22.1/include/linux/ext4_fs_extents.h
===================================================================
--- linux-2.6.22.1.orig/include/linux/ext4_fs_extents.h
+++ linux-2.6.22.1/include/linux/ext4_fs_extents.h
@@ -131,8 +131,8 @@ struct ext4_ext_path {
* callback must return valid extent (passed or newly created)
*/
typedef int (*ext_prepare_callback)(struct inode *, struct ext4_ext_path *,
- struct ext4_ext_cache *,
- void *);
+ struct ext4_ext_cache *,
+ struct ext4_extent *, void *);
#define EXT_CONTINUE 0
#define EXT_BREAK 1
Index: linux-2.6.22.1/fs/ext4/extents.c
===================================================================
--- linux-2.6.22.1.orig/fs/ext4/extents.c
+++ linux-2.6.22.1/fs/ext4/extents.c
@@ -42,7 +42,7 @@
#include <linux/falloc.h>
#include <linux/ext4_fs_extents.h>
#include <asm/uaccess.h>
-
+#include "fiemap.h"
/*
* ext_pblock:
@@ -1513,7 +1513,7 @@ int ext4_ext_walk_space(struct inode *in
}
BUG_ON(cbex.ec_len == 0);
- err = func(inode, path, &cbex, cbdata);
+ err = func(inode, path, &cbex, ex, cbdata);
ext4_ext_drop_refs(path);
if (err < 0)
@@ -2631,3 +2631,141 @@ retry:
return ret > 0 ? ret2 : ret;
}
+
+struct fiemap_internal {
+ struct fiemap *fiemap_s;
+ struct fiemap_extent fm_extent;
+ size_t tot_mapping_len;
+ char *cur_ext_ptr;
+ int current_extent;
+ int err;
+};
+
+ /*
+ * Callback function called for each extent to gather fiemap information.
+ */
+int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
+ struct ext4_ext_cache *newex, struct ext4_extent *ex,
+ void *data)
+{
+ struct fiemap_internal *fiemap_i = data;
+ struct fiemap *fiemap_s = fiemap_i->fiemap_s;
+ struct fiemap_extent *fm_extent = &fiemap_i->fm_extent;
+ int current_extent = fiemap_i->current_extent;
+ unsigned long blksize_bits = inode->i_sb->s_blocksize_bits;
+
+ /*
+ * ext4_ext_walk_space returns a hole for extents that have not been
+ * allocated yet.
+ */
+ if (((u64)(newex->ec_block + newex->ec_len) << blksize_bits >=
+ inode->i_size) && !ext4_ext_is_uninitialized(ex) &&
+ newex->ec_type == EXT4_EXT_CACHE_GAP)
+ return EXT_BREAK;
+
+ /*
+ * We only need to return number of extents.
+ */
+ if (fiemap_s->fm_flags & FIEMAP_FLAG_NUM_EXTENTS)
+ goto count_extents;
+
+ if (current_extent >= fiemap_s->fm_extent_count)
+ return EXT_BREAK;
+
+ memset(fm_extent, 0, sizeof(*fm_extent));
+ fm_extent->fe_offset = (__u64)newex->ec_start << blksize_bits;
+ fm_extent->fe_length = (__u64)newex->ec_len << blksize_bits;
+ fiemap_i->tot_mapping_len += fm_extent->fe_length;
+
+ if (newex->ec_type == EXT4_EXT_CACHE_GAP)
+ fm_extent->fe_flags |= FIEMAP_EXTENT_HOLE;
+
+ if (ext4_ext_is_uninitialized(ex))
+ fm_extent->fe_flags |= (FIEMAP_EXTENT_DELALLOC |
+ FIEMAP_EXTENT_UNMAPPED);
+
+ /*
+ * Mark this fiemap_extent as FIEMAP_EXTENT_EOF if it's past the end
+ * of file.
+ */
+ if ((u64)(newex->ec_block + newex->ec_len) << blksize_bits >=
+ inode->i_size)
+ fm_extent->fe_flags |= FIEMAP_EXTENT_EOF;
+
+ if (!copy_to_user(fiemap_i->cur_ext_ptr, fm_extent,
+ sizeof(struct fiemap_extent))) {
+ fiemap_i->cur_ext_ptr += sizeof(struct fiemap_extent);
+ } else {
+ fiemap_i->err = -EFAULT;
+ return EXT_BREAK;
+ }
+
+count_extents:
+ fiemap_i->current_extent++;
+
+ /*
+ * Stop if we are beyond requested mapping size but return complete last
+ * extent.
+ */
+ if ((u64)(newex->ec_block + newex->ec_len) << blksize_bits >=
+ fiemap_s->fm_length)
+ return EXT_BREAK;
+
+ return EXT_CONTINUE;
+}
+
+int ext4_fiemap(struct inode *inode, struct file *filp, unsigned int cmd,
+ unsigned long arg)
+{
+ struct fiemap *fiemap_s;
+ struct fiemap_internal fiemap_i;
+ struct fiemap_extent *last_extent;
+ ext4_fsblk_t start_blk;
+ int err = 0;
+
+ if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
+ return -EOPNOTSUPP;
+
+ fiemap_s = kmalloc(sizeof(*fiemap_s), GFP_KERNEL);
+ if (fiemap_s == NULL)
+ return -ENOMEM;
+ if (copy_from_user(fiemap_s, (struct fiemap __user *)arg,
+ sizeof(*fiemap_s)))
+ return -EFAULT;
+
+ if (fiemap_s->fm_flags & FIEMAP_FLAG_INCOMPAT)
+ return -EOPNOTSUPP;
+
+ if (fiemap_s->fm_flags & FIEMAP_FLAG_SYNC)
+ ext4_sync_file(filp, filp->f_path.dentry, 1);
+
+ start_blk = (fiemap_s->fm_start + inode->i_sb->s_blocksize - 1) >>
+ inode->i_sb->s_blocksize_bits;
+ fiemap_i.fiemap_s = fiemap_s;
+ fiemap_i.tot_mapping_len = 0;
+ fiemap_i.cur_ext_ptr = (char *)(arg + sizeof(*fiemap_s));
+ fiemap_i.current_extent = 0;
+ fiemap_i.err = 0;
+
+ /*
+ * Walk the extent tree gathering extent information
+ */
+ mutex_lock(&EXT4_I(inode)->truncate_mutex);
+ err = ext4_ext_walk_space(inode, start_blk , EXT_MAX_BLOCK - start_blk,
+ (void *)ext4_ext_fiemap_cb, &fiemap_i);
+ mutex_unlock(&EXT4_I(inode)->truncate_mutex);
+ if (err)
+ return err;
+
+ fiemap_s->fm_extent_count = fiemap_i.current_extent;
+ fiemap_s->fm_length = fiemap_i.tot_mapping_len;
+ if (fiemap_i.current_extent != 0 &&
+ !(fiemap_s->fm_flags & FIEMAP_FLAG_NUM_EXTENTS)) {
+ last_extent = &fiemap_i.fm_extent;
+ last_extent->fe_flags |= FIEMAP_EXTENT_LAST;
+ }
+ err = copy_to_user((void *)arg, fiemap_s, sizeof(*fiemap_s));
+
+ return err;
+}
+
Index: linux-2.6.22.1/fs/ext4/fiemap.h
===================================================================
--- /dev/null
+++ linux-2.6.22.1/fs/ext4/fiemap.h
@@ -0,0 +1,49 @@
+/*
+ * linux/fs/ext3/fiemap.h
+ *
+ * Copyright (C) 2007 Cluster File Systems, Inc
+ *
+ * Author: Kalpak Shah <kalpak@xxxxxxxxxxxxx>
+ */
+
+#ifndef _LINUX_EXT4_FIEMAP_H
+#define _LINUX_EXT4_FIEMAP_H
+
+struct fiemap_extent {
+ __u64 fe_offset; /* offset in bytes for the start of the extent */
+ __u64 fe_length; /* length in bytes for the extent */
+ __u32 fe_flags; /* returned FIEMAP_EXTENT_* flags for the extent */
+ __u32 fe_lun; /* logical device number for extent (starting at 0)*/
+};
+
+/*
+ * fiemap is not ext4-specific and should be moved into fs.h eventually.
+ */
+
+struct fiemap {
+ __u64 fm_start; /* logical starting byte offset (in/out) */
+ __u64 fm_length; /* logical length of map (in/out) */
+ __u32 fm_flags; /* FIEMAP_FLAG_* flags for request (in/out) */
+ __u32 fm_extent_count; /* number of extents in fm_extents (in/out) */
+ __u64 fm_unused;
+ struct fiemap_extent fm_extents[0];
+};
+
+#define FIEMAP_FLAG_SYNC 0x00000001 /* sync file data before map */
+#define FIEMAP_FLAG_HSM_READ 0x00000002 /* get data from HSM before map */
+#define FIEMAP_FLAG_NUM_EXTENTS 0x00000004 /* return only number of extents */
+#define FIEMAP_FLAG_INCOMPAT 0xff000000 /* error for unknown flags in here */
+
+#define FIEMAP_EXTENT_HOLE 0x00000001 /* has no data or space allocation */
+#define FIEMAP_EXTENT_UNWRITTEN 0x00000002 /* space allocated, but no data */
+#define FIEMAP_EXTENT_UNMAPPED 0x00000004 /* has data but no space allocation*/
+#define FIEMAP_EXTENT_ERROR 0x00000008 /* mapping error, errno in fe_start*/
+#define FIEMAP_EXTENT_NO_DIRECT 0x00000010 /* cannot access data directly */
+#define FIEMAP_EXTENT_LAST 0x00000020 /* last extent in the file */
+#define FIEMAP_EXTENT_DELALLOC 0x00000040 /* has data but not yet written,
+ * must have EXTENT_UNKNOWN set */
+#define FIEMAP_EXTENT_SECONDARY 0x00000080 /* data (also) in secondary storage,
+ * not in primary if EXTENT_UNKNOWN*/
+#define FIEMAP_EXTENT_EOF 0x00000100 /* if fm_start+fm_len is beyond EOF*/
+
+#endif /* _LINUX_EXT4_FIEMAP_H */
