Commit 14516bb7 authored by Dmitry Monakhov's avatar Dmitry Monakhov Committed by Theodore Ts'o
Browse files

ext4: fix suboptimal seek_{data,hole} extents traversial 

It is ridiculous practice to scan inode block by block, this technique
applicable only for old indirect files. This takes significant amount
of time for really large files. Let's reuse ext4_fiemap which already
traverse inode-tree in most optimal meaner.

TESTCASE:
ftruncate64(fd, 0);
ftruncate64(fd, 1ULL << 40);
/* lseek will spin very long time */
lseek64(fd, 0, SEEK_DATA);
lseek64(fd, 0, SEEK_HOLE);

Original report: https://lkml.org/lkml/2014/10/16/620



Signed-off-by: default avatarDmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: default avatarTheodore Ts'o <tytso@mit.edu>
parent d952d69e

fs/ext4/extents.c

+2 −2
Original line number Diff line number Diff line
@@ -5166,7 +5166,7 @@ int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, 


	/* fallback to generic here if not in extents fmt */

	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))

		return generic_block_fiemap(inode, fieinfo, start, len,

		return __generic_block_fiemap(inode, fieinfo, start, len,

					      ext4_get_block);



	if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))

fs/ext4/file.c

+106 −114
Original line number Diff line number Diff line
@@ -273,24 +273,19 @@ static int ext4_file_open(struct inode * inode, struct file * filp) 
 * we determine this extent as a data or a hole according to whether the

 * page cache has data or not.

 */

static int ext4_find_unwritten_pgoff(struct inode *inode,

				     int whence,

				     struct ext4_map_blocks *map,

				     loff_t *offset)

static int ext4_find_unwritten_pgoff(struct inode *inode, int whence,

				     loff_t endoff, loff_t *offset)

{

	struct pagevec pvec;

	unsigned int blkbits;

	pgoff_t index;

	pgoff_t end;

	loff_t endoff;

	loff_t startoff;

	loff_t lastoff;

	int found = 0;



	blkbits = inode->i_sb->s_blocksize_bits;

	startoff = *offset;

	lastoff = startoff;

	endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;





	index = startoff >> PAGE_CACHE_SHIFT;

	end = endoff >> PAGE_CACHE_SHIFT;
@@ -408,147 +403,144 @@ out: 
static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)

{

	struct inode *inode = file->f_mapping->host;

	struct ext4_map_blocks map;

	struct extent_status es;

	ext4_lblk_t start, last, end;

	loff_t dataoff, isize;

	int blkbits;

	int ret = 0;

	struct fiemap_extent_info fie;

	struct fiemap_extent ext[2];

	loff_t next;

	int i, ret = 0;



	mutex_lock(&inode->i_mutex);



	isize = i_size_read(inode);

	if (offset >= isize) {

	if (offset >= inode->i_size) {

		mutex_unlock(&inode->i_mutex);

		return -ENXIO;

	}

	fie.fi_flags = 0;

	fie.fi_extents_max = 2;

	fie.fi_extents_start = (struct fiemap_extent __user *) &ext;

	while (1) {

		mm_segment_t old_fs = get_fs();



	blkbits = inode->i_sb->s_blocksize_bits;

	start = offset >> blkbits;

	last = start;

	end = isize >> blkbits;

	dataoff = offset;

		fie.fi_extents_mapped = 0;

		memset(ext, 0, sizeof(*ext) * fie.fi_extents_max);



	do {

		map.m_lblk = last;

		map.m_len = end - last + 1;

		ret = ext4_map_blocks(NULL, inode, &map, 0);

		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {

			if (last != start)

				dataoff = (loff_t)last << blkbits;

		set_fs(get_ds());

		ret = ext4_fiemap(inode, &fie, offset, maxsize - offset);

		set_fs(old_fs);

		if (ret)

			break;



		/* No extents found, EOF */

		if (!fie.fi_extents_mapped) {

			ret = -ENXIO;

			break;

		}

		for (i = 0; i < fie.fi_extents_mapped; i++) {

			next = (loff_t)(ext[i].fe_length + ext[i].fe_logical);



			if (offset < (loff_t)ext[i].fe_logical)

				offset = (loff_t)ext[i].fe_logical;

			/*

		 * If there is a delay extent at this offset,

		 * it will be as a data.

			 * If extent is not unwritten, then it contains valid

			 * data, mapped or delayed.

			 */

		ext4_es_find_delayed_extent_range(inode, last, last, &es);

		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {

			if (last != start)

				dataoff = (loff_t)last << blkbits;

			break;

		}

			if (!(ext[i].fe_flags & FIEMAP_EXTENT_UNWRITTEN))

				goto out;



			/*

			 * If there is a unwritten extent at this offset,

			 * it will be as a data or a hole according to page

			 * cache that has data or not.

			 */

		if (map.m_flags & EXT4_MAP_UNWRITTEN) {

			int unwritten;

			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,

							      &map, &dataoff);

			if (unwritten)

				break;

		}



		last++;

		dataoff = (loff_t)last << blkbits;

	} while (last <= end);

			if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,

						      next, &offset))

				goto out;



			if (ext[i].fe_flags & FIEMAP_EXTENT_LAST) {

				ret = -ENXIO;

				goto out;

			}

			offset = next;

		}

	}

	if (offset > inode->i_size)

		offset = inode->i_size;

out:

	mutex_unlock(&inode->i_mutex);

	if (ret)

		return ret;



	if (dataoff > isize)

		return -ENXIO;



	return vfs_setpos(file, dataoff, maxsize);

	return vfs_setpos(file, offset, maxsize);

}



/*

 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.

 * ext4_seek_hole() retrieves the offset for SEEK_HOLE

 */

static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)

{

	struct inode *inode = file->f_mapping->host;

	struct ext4_map_blocks map;

	struct extent_status es;

	ext4_lblk_t start, last, end;

	loff_t holeoff, isize;

	int blkbits;

	int ret = 0;

	struct fiemap_extent_info fie;

	struct fiemap_extent ext[2];

	loff_t next;

	int i, ret = 0;



	mutex_lock(&inode->i_mutex);



	isize = i_size_read(inode);

	if (offset >= isize) {

	if (offset >= inode->i_size) {

		mutex_unlock(&inode->i_mutex);

		return -ENXIO;

	}



	blkbits = inode->i_sb->s_blocksize_bits;

	start = offset >> blkbits;

	last = start;

	end = isize >> blkbits;

	holeoff = offset;

	fie.fi_flags = 0;

	fie.fi_extents_max = 2;

	fie.fi_extents_start = (struct fiemap_extent __user *)&ext;

	while (1) {

		mm_segment_t old_fs = get_fs();



	do {

		map.m_lblk = last;

		map.m_len = end - last + 1;

		ret = ext4_map_blocks(NULL, inode, &map, 0);

		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {

			last += ret;

			holeoff = (loff_t)last << blkbits;

			continue;

		}

		fie.fi_extents_mapped = 0;

		memset(ext, 0, sizeof(*ext));



		set_fs(get_ds());

		ret = ext4_fiemap(inode, &fie, offset, maxsize - offset);

		set_fs(old_fs);

		if (ret)

			break;



		/* No extents found */

		if (!fie.fi_extents_mapped)

			break;



		for (i = 0; i < fie.fi_extents_mapped; i++) {

			next = (loff_t)(ext[i].fe_logical + ext[i].fe_length);

			/*

		 * If there is a delay extent at this offset,

		 * we will skip this extent.

			 * If extent is not unwritten, then it contains valid

			 * data, mapped or delayed.

			 */

		ext4_es_find_delayed_extent_range(inode, last, last, &es);

		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {

			last = es.es_lblk + es.es_len;

			holeoff = (loff_t)last << blkbits;

			if (!(ext[i].fe_flags & FIEMAP_EXTENT_UNWRITTEN)) {

				if (offset < (loff_t)ext[i].fe_logical)

					goto out;

				offset = next;

				continue;

			}



			/*

			 * If there is a unwritten extent at this offset,

			 * it will be as a data or a hole according to page

			 * cache that has data or not.

			 */

		if (map.m_flags & EXT4_MAP_UNWRITTEN) {

			int unwritten;

			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,

							      &map, &holeoff);

			if (!unwritten) {

				last += ret;

				holeoff = (loff_t)last << blkbits;

				continue;

			if (ext4_find_unwritten_pgoff(inode, SEEK_HOLE,

						      next, &offset))

				goto out;



			offset = next;

			if (ext[i].fe_flags & FIEMAP_EXTENT_LAST)

				goto out;

		}

	}



		/* find a hole */

		break;

	} while (last <= end);



	if (offset > inode->i_size)

		offset = inode->i_size;

out:

	mutex_unlock(&inode->i_mutex);

	if (ret)

		return ret;



	if (holeoff > isize)

		holeoff = isize;



	return vfs_setpos(file, holeoff, maxsize);

	return vfs_setpos(file, offset, maxsize);

}



/*

CRA Git | Maintained and supported by SUSTech CRA and CCSE