Finding File Types (Unix Power Tools, 3rd Edition)

start page | rating of books | rating of authors | reviews | copyrights

12.6. Finding File Types

Many different kinds of files live on the typical Unix system: database files, executable files, regular text files, files for applications like StarOffice, tar files, mail messages, directories, font files, and so on.

You often want to check to make sure you have the right "kind" of file before doing something. For example, you'd like to read the file tar. But before typing more tar, you'd like to know whether this file is your set of notes on carbon-based sludge or the tar executable. If you're wrong, the consequences might be unpleasant. Sending the tar executable to your screen might screw up your terminal settings, log you off, or do any number of unpleasant things.

Go to http://examples.oreilly.com/upt3 for more information on: file

The file utility tells you what sort of file something is. It's fairly self-explanatory:

% file /bin/sh
/bin/sh:       sparc demand paged executable
% file 2650
2650:          [nt]roff, tbl, or eqn input text
% file 0001,v
0001,v:        ascii text
% file foo.sh
foo.sh:        shell commands

file is actually quite clever though it isn't always correct -- some versions are better than others. It doesn't just tell you if something's binary or text; it looks at the beginning of the file and tries to figure out what it's doing. So, for example, you see that file 2650 is an nroff (Section 45.12) file and foo.sh is a shell script. It isn't quite clever enough to figure out that 0001,v is an RCS (Section 39.5) archive, but it does know that it's a plain ASCII text file.

Many versions of file can be customized to recognize additional file types. The file /etc/magic tells file how to recognize different kinds of files. [My Linux system has the file command from ftp://ftp.astron.com/pub/file/, which uses a multiple-database format. It's updated fairly often to understand new file formats. -- JP] It's capable of a lot (and should be capable of even more), but we'll satisfy ourselves with an introductory explanation. Our goal will be to teach file to recognize RCS archives.

/etc/magic has four fields:

offset data-type value file-type

These are as follows:

offset: The offset into the file at which magic will try to find something. If you're looking for something right at the beginning of the file, the offset should be 0. (This is usually what you want.)
data-type: The type of test to make. Use string for text comparisons, byte for byte comparisons, short for two-byte comparisons, and long for four-byte comparisons.
value: The value you want to find. For string comparisons, any text string will do; you can use the standard Unix escape sequences (such as \n for newline). For numeric comparisons (byte, short, long), this field should be a number, expressed as a C constant (e.g., 0x77 for the hexadecimal byte 77).
file-type: The string that file will print if this test succeeds.

So, we know that RCS archives begin with the word head. This word is right at the beginning of the file (offset 0). Since we obviously want a string comparison, we make the the following addition to /etc/magic:

0     string     head     RCS archive

This says, "The file is an RCS archive if you find the string head at an offset of 0 bytes from the beginning of the file." Does it work?

% file RCS/0001,v
RCS/0001,v:        RCS archive

As I said, the tests can be much more complicated, particularly if you're working with binary files. To recognize simple text files, this is all you need to know.

-- ML


12.5. What's in That Whitespace?		12.7. Squash Extra Blank Lines