Can a file that was originally sparse and then expanded be made sparse again

filessparse-filesUtilities

I know that copying or transferring what was originally a sparse file without using a utility that understands sparse files will cause the 'holes' to be filled out. Is there a method or utility to turn what was once a sparse file back to sparse?

For Example:
create sparse file:

% dd if=/dev/zero of=TEST bs=1 count=0 seek=1G
# do some op that pads out the holes
% scp TEST localhost:~/TEST2
% ls -lhs TEST*
   0 -rw-rw-r--. 1 tony tony 1.0G Oct 16 13:35 TEST
1.1G -rw-rw-r--. 1 tony tony 1.0G Oct 16 13:37 TEST2

Is there some way to:

% resparse TEST2
to get:
   0 -rw-rw-r--. 1 tony tony 1.0G Oct 16 13:35 TEST
  0G -rw-rw-r--. 1 tony tony 1.0G Oct 16 13:37 TEST2

Best Answer

Edit 2015

as of util-linux 2.25, the fallocate utility on Linux has a -d/--dig-hole option for that.

fallocate -d the-file

Would dig a hole for every block full of zeros in the file


On older systems, you can do it by hand:

Linux has a FALLOC_FL_PUNCH_HOLE option to fallocate that can do this. I found a script on github with an example:

Using FALLOC_FL_PUNCH_HOLE from Python

I modified it a bit to do what you asked -- punch holes in regions of files that are filled with zeros. Here it is:

Using FALLOC_FL_PUNCH_HOLE from Python to punch holes in files

usage: punch.py [-h] [-v VERBOSE] FILE [FILE ...]

Punch out the empty areas in a file, making it sparse

positional arguments:
  FILE                  file(s) to modify in-place

optional arguments:
  -h, --help            show this help message and exit
  -v VERBOSE, --verbose VERBOSE
                        be verbose

Example:

# create a file with some data, a hole, and some more data
$ dd if=/dev/urandom of=test1 bs=4096 count=1 seek=0
$ dd if=/dev/urandom of=test1 bs=4096 count=1 seek=2

# see that it has holes
$ du --block-size=1 --apparent-size test1
12288   test1
$ du --block-size=1 test1
8192    test1

# copy it, ignoring the hole
$ cat test1 > test2
$ du --block-size=1 --apparent-size test2
12288   test2
$ du --block-size=1 test2
12288    test2

# punch holes again
$ ./punch.py test2
$ du --block-size=1 --apparent-size test2
12288   test2
$ du --block-size=1 test2
8192    test2

# verify
$ cmp test1 test2 && echo "files are the same"
files are the same

Note that punch.py only finds blocks of 4096 bytes to punch out, so it might not make a file exactly as sparse as it was when you started. It could be made smarter, of course. Also, it's only lightly tested, so be careful and make backups before trusting it!

Related Question