Detecting type of file in a program

[Guy Harris]

>Just what are magic numbers for

Some software subsystems mark files of some format that they know about
with some particular number or set of characters, usually at the front
of the file.  The name, at least as I remember it being used in UNIX
systems, originally referred to stuff stuck at the end of object and
executable files that flagged the type of executable file.

Back in the dark ages, I think an executable file began with a PDP-11
branch instruction that jumped around a header that specified the size
of the code+data in the file; the instruction was a 16-bit number, value
octal 407.  When separate text&data, and split I&D, executables were
added, "407" as the first 16 bits of the executable file was treated as
a flag indicating that the image was a non-separate text&data image, and
410 was used for separate text&data and 411 (? - I haven't dealt with
PDP-11s in ages...) meant separate I&D.  (Also, the header was no longer
part of the image itself, since 410 and 411 would branch further than
407 did and I guess they thought it was silly to pad the header to cope
with this...).

Anyway, the term "magic number" then applied to other flags stuck at the
front of files by subsystems; e.g, 0177555 (word size) for a very old archive
format, 0177545 (16 bits) for an older archive format, "!<arch>\n" as a
string for the "modern" archive format (4.xBSD/S5R2 and later), etc.

>and of what significance is the /etc/magic file.

The S5 version of the "file" command (S5R2 and later, anyway) has a file
"/etc/magic" that it can read to find out some of the magic
numbers/strings it should look for; that way, you can extend its
repertoire without having to hack the source code.  SunOS 3.2 and later
use a "file" based on this, with some extensions.

>Every once in a while I get an error saying something about bad
>magic numbers or whatnot, and I usually just punt.  What's the scoop?

That probably means you tried to link or execute a file that the linker
or the "exec" call didn't recognize as an object or executable file,
because it didn't have a proper magic number at the front of it; hence,
"bad magic number".