In computer science, a sparse file is a type of computer file that attempts to use file system space more efficiently when the file itself is partially empty. This is achieved by writing brief information (metadata) representing the empty blocks to disk instead of the actual "empty" space which makes up the block, using less disk space. The full block size is written to disk as the actual size only when the block contains "real" (non-empty) data.
When reading sparse files, the file system transparently converts metadata representing empty blocks into "real" blocks filled with null bytes at runtime. The application is unaware of this conversion.
Most modern file systems support sparse files, including most Unix variants and NTFS. Apple's HFS+ does not provide for sparse files, but in OS X, the virtual file system layer supports storing them in any supported file system, including HFS+. Apple File System (APFS), announced in June 2016 at WWDC, also supports them. Sparse files are commonly used for disk images, database snapshots, log files and in scientific applications.
The advantage of sparse files is that storage is only allocated when actually needed: disk space is saved, and large files can be created even if there is insufficient free space on the file system. This also reduces the time of the first write as the system doesn't have to allocate blocks for the "skipped" space. If the initial allocation requires writing all zeros to the space, it also keeps the system from having to write over the "skipped" space twice.
For example, a virtual machine image with max size of 100GB that has 2GB of files actually written would require the full 100GB when backed by pre-allocated storage, yet only 2GB on a sparse file. If the file system supports hole punching and the guest operating system issues TRIM commands, deleting files on the guest will accordingly reduce space needed.
Disadvantages are that sparse files may become fragmented; file system free space reports may be misleading; filling up file systems containing sparse files can have unexpected effects (such as disk-full or quota-exceeded errors when merely overwriting an existing portion of a file that happened to have been sparse); and copying a sparse file with a program that does not explicitly support them may copy the entire, uncompressed size of the file, including the zero sections which are not allocated on disk—losing the benefits of the sparse property in the file. Sparse files are also not fully supported by all backup software or applications. However, the VFS implementation sidesteps the prior two disadvantages. Loading executables on 32bit Windows (exe or dll) which are sparse takes a much longer time, since the file cannot be memory mapped in the limited 4GB address space, and are not cached as there is no codepath for caching 32bit sparse executables (Windows on 64bit architectures can map sparse executables). On NTFS sparse file (or rather its non-zero areas) can't be compressed. NTFS implements sparseness as a special kind of compression so a file may be either sparse or compressed.
Sparse files are typically handled transparently to the user. But the differences between a normal file and sparse file become apparent in some situations.
The Unix command
dd of=sparse-file bs=5M seek=1 count=0
will create a file of five mebibytes in size, but with no data stored on disk (only metadata). (GNU
dd has this behavior because it calls
ftruncate to set the file size; other implementations may merely create an empty file.)
Similarly the truncate command may be used, if available:
truncate -s 5M <filename>
On Linux, an existing file can be converted to sparse by:
fallocate -d <filename>
Alas, there's no portable way to punch holes; the syscall is fallocate(FALLOC_FL_PUNCH_HOLE) on Linux, fcntl(F_FREESP) on Solaris.
-s option of the
ls command shows the occupied space in blocks.
ls -ls sparse-file
du command prints the occupied space, while
ls prints the apparent size.
In some non-standard versions of
du, the option
--block-size=1 prints the occupied space in bytes instead of blocks, so that it can be compared to the
du --block-size=1 sparse-file ls -l sparse-file
Normally, the GNU version of
cp is good at detecting whether a file is sparse, so
cp sparse-file new-file
creates new-file, which will be sparse. However, GNU cp does have a
--sparse option. This is especially useful if a file containing long zero blocks is saved in a non-sparse way (i.e. the zero blocks have been written out to disk in full). Disk space can be saved by doing:
cp --sparse=always file1 file1_sparsed
Some cp implementations, like FreeBSD's cp, do not support the
--sparse option and will always expand sparse files. A partially viable alternative on those systems is to use rsync with its own
--sparse option instead of cp. Unfortunately
--sparse cannot be combined with
cp --sparse=always /proc/self/fd/0 new-sparse-file < somefile