Is there any commandline tools to determine filesystem type on a block device before mounting it?and how is that achieved?

I believe it is possible since I usually mount external disks with

$ mount /dev/sdXX /mnt

mount automatically determines the filesystem for you. modern GUI tools even probe for disk usage and other info without mounting the filesystem if the driver for that fs is present.

the scenario here is that

  • the partition type and filesystem type may mismatch.
  • most linux filesystem use partition type "83", which doesn't offer much info about the fs it contains.
  • the corresponding drive may be absent, missing xfsprogs, hfsprogs, etc.

when auto-mounting fails, with an arbitrary partition or disk image in hand, it's simply not feasible to try each fs type candidate until you find the right one. or what if the filesystem is corrupt. you can't diagnose it with the designate tool as it's fs type is unknown.

i think superblock is where most filesystem stores its identifier. but different fs write superblock at different places.

is the raw dump of data of XXX bytes in the beginning sufficient to determine the fs type? is there a standard on where and in what format one should store such info?

any insight on this issue is much appreciated.



thanx for philag's answer. so the usual file approach is actually the best approach.

my problem was just that i encountered a weird filesystem, whose file output is rather useless.

digging deeper into file's documentation (maybe the system calls it utilizes as well) should help me understand this issue better.

  • 1
    In response to the update: file just reads the file, it doesn't use any system calls except those you need to read the file.
    – phihag
    Commented Aug 28, 2011 at 12:12

1 Answer 1


The first bytes (not literally, but usually in the first 4KiB) contain a signature, which especially crafted to be unique. The file utility can determine these signatures. See for yourself:

$ # Create an example file we can write to. vdisk stands for your partition.
$ dd if=/dev/zero of=vdisk bs=1M count=40
$ mkfs.ext2 -qF vdisk
$ file vdisk
vdisk: Linux rev 1.0 ext2 filesystem data, UUID=cce25572-...-f4eba2957279
$ mkfs.xfs -fq vdisk
$ file vdisk
vdisk: SGI XFS filesystem data (blksz 4096, inosz 256, v2 dirs)
$ # How does file find out? Let's look inside the partition
$ hexdump vdisk -C  | head -n 1
00000000  58 46 53 42 00 00 10 00  00 00 00 00 00 00 28 00  |XFSB..........(.|

To get a detailed list of filesystems that file recognizes, have a look at magic/Magdir/filesystems in the file source code. If you're just interested in those supported by your kernel, inspect include/linux/magic.h in your kernel sources.

If you are interested in a block device special the -s option is also useful, as is -L that will follow (dereference) symbolic links rather than operate on the link itself. For example if using a logical volume and device-mapper something like:

file -Ls /dev/mapper/home

may be useful.

  • i understand that, i use that to determine disk images. and you have to dd the first bytes to a image file to run file on it, or file always shows "block special" for block devices.
    – Huang Tao
    Commented Aug 28, 2011 at 11:23
  • but it's not guaranteed to work right? file has to keep a record of what signature maps to what filesystem.
    – Huang Tao
    Commented Aug 28, 2011 at 11:25
  • @Huang Tao Sorry, I didn't read your question fully and therefore answered at the wrong level. Use the -s option (as in file -s /dev/sda1 ) to examine block devices.
    – phihag
    Commented Aug 28, 2011 at 11:26
  • @Huang Tao Well, it works good enough to distinguish all the file systems supported by the kernel. Most good disk formats (file systems as well as files) include a precisely crafted header to make them differ from any other file type.
    – phihag
    Commented Aug 28, 2011 at 11:28
  • 1
    of course i was using /dev/sdb1, not /dev/sdb. from what i can tell, the partition is pretending to be a disk.
    – Huang Tao
    Commented Aug 28, 2011 at 11:35

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .