xref: /freebsd/contrib/file/doc/magic.man (revision d3d381b2b194b4d24853e92eecef55f262688d1a)
1.It S2
2.\" $File: magic.man,v 1.93 2018/06/22 20:39:49 christos Exp $
3.Dd June 22, 2018
4.Dt MAGIC __FSECTION__
5.Os
6.\" install as magic.4 on USG, magic.5 on V7, Berkeley and Linux systems.
7.Sh NAME
8.Nm magic
9.Nd file command's magic pattern file
10.Sh DESCRIPTION
11This manual page documents the format of magic files as
12used by the
13.Xr file __CSECTION__
14command, version __VERSION__.
15The
16.Xr file __CSECTION__
17command identifies the type of a file using,
18among other tests,
19a test for whether the file contains certain
20.Dq "magic patterns" .
21The database of these
22.Dq "magic patterns"
23is usually located in a binary file in
24.Pa __MAGIC__.mgc
25or a directory of source text magic pattern fragment files in
26.Pa __MAGIC__ .
27The database specifies what patterns are to be tested for, what message or
28MIME type to print if a particular pattern is found,
29and additional information to extract from the file.
30.Pp
31The format of the source fragment files that are used to build this database
32is as follows:
33Each line of a fragment file specifies a test to be performed.
34A test compares the data starting at a particular offset
35in the file with a byte value, a string or a numeric value.
36If the test succeeds, a message is printed.
37The line consists of the following fields:
38.Bl -tag -width ".Dv message"
39.It Dv offset
40A number specifying the offset (in bytes) into the file of the data
41which is to be tested.
42This offset can be a negative number if it is:
43.Bl -bullet  -compact
44.It
45The first direct offset of the magic entry (at continuation level 0),
46in which case it is interpreted an offset from end end of the file
47going backwards.
48This works only when a file descriptor to the file is a available and it
49is a regular file.
50.It
51A continuation offset relative to the end of the last up-level field
52.Dv ( \*[Am] ) .
53.El
54.It Dv type
55The type of the data to be tested.
56The possible values are:
57.Bl -tag -width ".Dv lestring16"
58.It Dv byte
59A one-byte value.
60.It Dv short
61A two-byte value in this machine's native byte order.
62.It Dv long
63A four-byte value in this machine's native byte order.
64.It Dv quad
65An eight-byte value in this machine's native byte order.
66.It Dv float
67A 32-bit single precision IEEE floating point number in this machine's native byte order.
68.It Dv double
69A 64-bit double precision IEEE floating point number in this machine's native byte order.
70.It Dv string
71A string of bytes.
72The string type specification can be optionally followed
73by /[WwcCtbT]*.
74The
75.Dq W
76flag compacts whitespace in the target, which must
77contain at least one whitespace character.
78If the magic has
79.Dv n
80consecutive blanks, the target needs at least
81.Dv n
82consecutive blanks to match.
83The
84.Dq w
85flag treats every blank in the magic as an optional blank.
86The
87.Dq c
88flag specifies case insensitive matching: lower case
89characters in the magic match both lower and upper case characters in the
90target, whereas upper case characters in the magic only match upper case
91characters in the target.
92The
93.Dq C
94flag specifies case insensitive matching: upper case
95characters in the magic match both lower and upper case characters in the
96target, whereas lower case characters in the magic only match upper case
97characters in the target.
98To do a complete case insensitive match, specify both
99.Dq c
100and
101.Dq C .
102The
103.Dq t
104flag forces the test to be done for text files, while the
105.Dq b
106flag forces the test to be done for binary files.
107The
108.Dq T
109flag causes the string to be trimmed, i.e. leading and trailing whitespace
110is deleted before the string is printed.
111.It Dv pstring
112A Pascal-style string where the first byte/short/int is interpreted as the
113unsigned length.
114The length defaults to byte and can be specified as a modifier.
115The following modifiers are supported:
116.Bl -tag -compact -width B
117.It B
118A byte length (default).
119.It H
120A 4 byte big endian length.
121.It h
122A 2 byte big endian length.
123.It L
124A 4 byte little endian length.
125.It l
126A 2 byte little endian length.
127.It J
128The length includes itself in its count.
129.El
130The string is not NUL terminated.
131.Dq J
132is used rather than the more
133valuable
134.Dq I
135because this type of length is a feature of the JPEG
136format.
137.It Dv date
138A four-byte value interpreted as a UNIX date.
139.It Dv qdate
140A eight-byte value interpreted as a UNIX date.
141.It Dv ldate
142A four-byte value interpreted as a UNIX-style date, but interpreted as
143local time rather than UTC.
144.It Dv qldate
145An eight-byte value interpreted as a UNIX-style date, but interpreted as
146local time rather than UTC.
147.It Dv qwdate
148An eight-byte value interpreted as a Windows-style date.
149.It Dv beid3
150A 32-bit ID3 length in big-endian byte order.
151.It Dv beshort
152A two-byte value in big-endian byte order.
153.It Dv belong
154A four-byte value in big-endian byte order.
155.It Dv bequad
156An eight-byte value in big-endian byte order.
157.It Dv befloat
158A 32-bit single precision IEEE floating point number in big-endian byte order.
159.It Dv bedouble
160A 64-bit double precision IEEE floating point number in big-endian byte order.
161.It Dv bedate
162A four-byte value in big-endian byte order,
163interpreted as a Unix date.
164.It Dv beqdate
165An eight-byte value in big-endian byte order,
166interpreted as a Unix date.
167.It Dv beldate
168A four-byte value in big-endian byte order,
169interpreted as a UNIX-style date, but interpreted as local time rather
170than UTC.
171.It Dv beqldate
172An eight-byte value in big-endian byte order,
173interpreted as a UNIX-style date, but interpreted as local time rather
174than UTC.
175.It Dv beqwdate
176An eight-byte value in big-endian byte order,
177interpreted as a Windows-style date.
178.It Dv bestring16
179A two-byte unicode (UCS16) string in big-endian byte order.
180.It Dv leid3
181A 32-bit ID3 length in little-endian byte order.
182.It Dv leshort
183A two-byte value in little-endian byte order.
184.It Dv lelong
185A four-byte value in little-endian byte order.
186.It Dv lequad
187An eight-byte value in little-endian byte order.
188.It Dv lefloat
189A 32-bit single precision IEEE floating point number in little-endian byte order.
190.It Dv ledouble
191A 64-bit double precision IEEE floating point number in little-endian byte order.
192.It Dv ledate
193A four-byte value in little-endian byte order,
194interpreted as a UNIX date.
195.It Dv leqdate
196An eight-byte value in little-endian byte order,
197interpreted as a UNIX date.
198.It Dv leldate
199A four-byte value in little-endian byte order,
200interpreted as a UNIX-style date, but interpreted as local time rather
201than UTC.
202.It Dv leqldate
203An eight-byte value in little-endian byte order,
204interpreted as a UNIX-style date, but interpreted as local time rather
205than UTC.
206.It Dv leqwdate
207An eight-byte value in little-endian byte order,
208interpreted as a Windows-style date.
209.It Dv lestring16
210A two-byte unicode (UCS16) string in little-endian byte order.
211.It Dv melong
212A four-byte value in middle-endian (PDP-11) byte order.
213.It Dv medate
214A four-byte value in middle-endian (PDP-11) byte order,
215interpreted as a UNIX date.
216.It Dv meldate
217A four-byte value in middle-endian (PDP-11) byte order,
218interpreted as a UNIX-style date, but interpreted as local time rather
219than UTC.
220.It Dv indirect
221Starting at the given offset, consult the magic database again.
222The offset of the
223.Dv indirect
224magic is by default absolute in the file, but one can specify
225.Dv /r
226to indicate that the offset is relative from the beginning of the entry.
227.It Dv name
228Define a
229.Dq named
230magic instance that can be called from another
231.Dv use
232magic entry, like a subroutine call.
233Named instance direct magic offsets are relative to the offset of the
234previous matched entry, but indirect offsets are relative to the beginning
235of the file as usual.
236Named magic entries always match.
237.It Dv use
238Recursively call the named magic starting from the current offset.
239If the name of the referenced begins with a
240.Dv ^
241then the endianness of the magic is switched; if the magic mentioned
242.Dv leshort
243for example,
244it is treated as
245.Dv beshort
246and vice versa.
247This is useful to avoid duplicating the rules for different endianness.
248.It Dv regex
249A regular expression match in extended POSIX regular expression syntax
250(like egrep).
251Regular expressions can take exponential time to process, and their
252performance is hard to predict, so their use is discouraged.
253When used in production environments, their performance
254should be carefully checked.
255The size of the string to search should also be limited by specifying
256.Dv /<length> ,
257to avoid performance issues scanning long files.
258The type specification can also be optionally followed by
259.Dv /[c][s][l] .
260The
261.Dq c
262flag makes the match case insensitive, while the
263.Dq s
264flag update the offset to the start offset of the match, rather than the end.
265The
266.Dq l
267modifier, changes the limit of length to mean number of lines instead of a
268byte count.
269Lines are delimited by the platforms native line delimiter.
270When a line count is specified, an implicit byte count also computed assuming
271each line is 80 characters long.
272If neither a byte or line count is specified, the search is limited automatically
273to 8KiB.
274.Dv ^
275and
276.Dv $
277match the beginning and end of individual lines, respectively,
278not beginning and end of file.
279.It Dv search
280A literal string search starting at the given offset.
281The same modifier flags can be used as for string patterns.
282The search expression must contain the range in the form
283.Dv /number,
284that is the number of positions at which the match will be
285attempted, starting from the start offset.
286This is suitable for
287searching larger binary expressions with variable offsets, using
288.Dv \e
289escapes for special characters.
290The order of modifier and number is not relevant.
291.It Dv default
292This is intended to be used with the test
293.Em x
294(which is always true) and it has no type.
295It matches when no other test at that continuation level has matched before.
296Clearing that matched tests for a continuation level, can be done using the
297.Dv clear
298test.
299.It Dv clear
300This test is always true and clears the match flag for that continuation level.
301It is intended to be used with the
302.Dv default
303test.
304.El
305.Pp
306For compatibility with the Single
307.Ux
308Standard, the type specifiers
309.Dv dC
310and
311.Dv d1
312are equivalent to
313.Dv byte ,
314the type specifiers
315.Dv uC
316and
317.Dv u1
318are equivalent to
319.Dv ubyte ,
320the type specifiers
321.Dv dS
322and
323.Dv d2
324are equivalent to
325.Dv short ,
326the type specifiers
327.Dv uS
328and
329.Dv u2
330are equivalent to
331.Dv ushort ,
332the type specifiers
333.Dv dI ,
334.Dv dL ,
335and
336.Dv d4
337are equivalent to
338.Dv long ,
339the type specifiers
340.Dv uI ,
341.Dv uL ,
342and
343.Dv u4
344are equivalent to
345.Dv ulong ,
346the type specifier
347.Dv d8
348is equivalent to
349.Dv quad ,
350the type specifier
351.Dv u8
352is equivalent to
353.Dv uquad ,
354and the type specifier
355.Dv s
356is equivalent to
357.Dv string .
358In addition, the type specifier
359.Dv dQ
360is equivalent to
361.Dv quad
362and the type specifier
363.Dv uQ
364is equivalent to
365.Dv uquad .
366.Pp
367Each top-level magic pattern (see below for an explanation of levels)
368is classified as text or binary according to the types used.
369Types
370.Dq regex
371and
372.Dq search
373are classified as text tests, unless non-printable characters are used
374in the pattern.
375All other tests are classified as binary.
376A top-level
377pattern is considered to be a test text when all its patterns are text
378patterns; otherwise, it is considered to be a binary pattern.
379When
380matching a file, binary patterns are tried first; if no match is
381found, and the file looks like text, then its encoding is determined
382and the text patterns are tried.
383.Pp
384The numeric types may optionally be followed by
385.Dv \*[Am]
386and a numeric value,
387to specify that the value is to be AND'ed with the
388numeric value before any comparisons are done.
389Prepending a
390.Dv u
391to the type indicates that ordered comparisons should be unsigned.
392.It Dv test
393The value to be compared with the value from the file.
394If the type is
395numeric, this value
396is specified in C form; if it is a string, it is specified as a C string
397with the usual escapes permitted (e.g. \en for new-line).
398.Pp
399Numeric values
400may be preceded by a character indicating the operation to be performed.
401It may be
402.Dv = ,
403to specify that the value from the file must equal the specified value,
404.Dv \*[Lt] ,
405to specify that the value from the file must be less than the specified
406value,
407.Dv \*[Gt] ,
408to specify that the value from the file must be greater than the specified
409value,
410.Dv \*[Am] ,
411to specify that the value from the file must have set all of the bits
412that are set in the specified value,
413.Dv ^ ,
414to specify that the value from the file must have clear any of the bits
415that are set in the specified value, or
416.Dv ~ ,
417the value specified after is negated before tested.
418.Dv x ,
419to specify that any value will match.
420If the character is omitted, it is assumed to be
421.Dv = .
422Operators
423.Dv \*[Am] ,
424.Dv ^ ,
425and
426.Dv ~
427don't work with floats and doubles.
428The operator
429.Dv !\&
430specifies that the line matches if the test does
431.Em not
432succeed.
433.Pp
434Numeric values are specified in C form; e.g.
435.Dv 13
436is decimal,
437.Dv 013
438is octal, and
439.Dv 0x13
440is hexadecimal.
441.Pp
442Numeric operations are not performed on date types, instead the numeric
443value is interpreted as an offset.
444.Pp
445For string values, the string from the
446file must match the specified string.
447The operators
448.Dv = ,
449.Dv \*[Lt]
450and
451.Dv \*[Gt]
452(but not
453.Dv \*[Am] )
454can be applied to strings.
455The length used for matching is that of the string argument
456in the magic file.
457This means that a line can match any non-empty string (usually used to
458then print the string), with
459.Em \*[Gt]\e0
460(because all non-empty strings are greater than the empty string).
461.Pp
462Dates are treated as numerical values in the respective internal
463representation.
464.Pp
465The special test
466.Em x
467always evaluates to true.
468.It Dv message
469The message to be printed if the comparison succeeds.
470If the string contains a
471.Xr printf 3
472format specification, the value from the file (with any specified masking
473performed) is printed using the message as the format string.
474If the string begins with
475.Dq \eb ,
476the message printed is the remainder of the string with no whitespace
477added before it: multiple matches are normally separated by a single
478space.
479.El
480.Pp
481An APPLE 4+4 character APPLE creator and type can be specified as:
482.Bd -literal -offset indent
483!:apple	CREATYPE
484.Ed
485.Pp
486A MIME type is given on a separate line, which must be the next
487non-blank or comment line after the magic line that identifies the
488file type, and has the following format:
489.Bd -literal -offset indent
490!:mime	MIMETYPE
491.Ed
492.Pp
493i.e. the literal string
494.Dq !:mime
495followed by the MIME type.
496.Pp
497An optional strength can be supplied on a separate line which refers to
498the current magic description using the following format:
499.Bd -literal -offset indent
500!:strength OP VALUE
501.Ed
502.Pp
503The operand
504.Dv OP
505can be:
506.Dv + ,
507.Dv - ,
508.Dv * ,
509or
510.Dv /
511and
512.Dv VALUE
513is a constant between 0 and 255.
514This constant is applied using the specified operand
515to the currently computed default magic strength.
516.Pp
517Some file formats contain additional information which is to be printed
518along with the file type or need additional tests to determine the true
519file type.
520These additional tests are introduced by one or more
521.Em \*[Gt]
522characters preceding the offset.
523The number of
524.Em \*[Gt]
525on the line indicates the level of the test; a line with no
526.Em \*[Gt]
527at the beginning is considered to be at level 0.
528Tests are arranged in a tree-like hierarchy:
529if the test on a line at level
530.Em n
531succeeds, all following tests at level
532.Em n+1
533are performed, and the messages printed if the tests succeed, until a line
534with level
535.Em n
536(or less) appears.
537For more complex files, one can use empty messages to get just the
538"if/then" effect, in the following way:
539.Bd -literal -offset indent
5400      string   MZ
541\*[Gt]0x18  leshort  \*[Lt]0x40   MS-DOS executable
542\*[Gt]0x18  leshort  \*[Gt]0x3f   extended PC executable (e.g., MS Windows)
543.Ed
544.Pp
545Offsets do not need to be constant, but can also be read from the file
546being examined.
547If the first character following the last
548.Em \*[Gt]
549is a
550.Em \&(
551then the string after the parenthesis is interpreted as an indirect offset.
552That means that the number after the parenthesis is used as an offset in
553the file.
554The value at that offset is read, and is used again as an offset
555in the file.
556Indirect offsets are of the form:
557.Em (( x [[.,][bBcCeEfFgGhHiIlmsSqQ]][+\-][ y ]) .
558The value of
559.Em x
560is used as an offset in the file.
561A byte, id3 length, short or long is read at that offset depending on the
562.Em [bBcCeEfFgGhHiIlmsSqQ]
563type specifier.
564The value is treated as signed if
565.Dq ,
566is specified or unsigned if
567.Dq .
568is specified.
569The capitalized types interpret the number as a big endian
570value, whereas the small letter versions interpret the number as a little
571endian value;
572the
573.Em m
574type interprets the number as a middle endian (PDP-11) value.
575To that number the value of
576.Em y
577is added and the result is used as an offset in the file.
578The default type if one is not specified is long.
579The following types are recognized:
580.Bl -column -offset indent "Type" "Half/Short" "Little" "Size"
581.It Sy Type	Sy Mnemonic	Sy Endian	Sy Size
582.It bcBc	Byte/Char	N/A	1
583.It efg	Double	Little	8
584.It EFG	Double	Big	8
585.It hs	Half/Short	Little	2
586.It HS	Half/Short	Big	2
587.It i	ID3	Little	4
588.It I	ID3	Big	4
589.It m	Middle	Middle	4
590.It q	Quad	Little	8
591.It Q	Quad	Big	8
592.El
593.Pp
594That way variable length structures can be examined:
595.Bd -literal -offset indent
596# MS Windows executables are also valid MS-DOS executables
5970           string  MZ
598\*[Gt]0x18       leshort \*[Lt]0x40   MZ executable (MS-DOS)
599# skip the whole block below if it is not an extended executable
600\*[Gt]0x18       leshort \*[Gt]0x3f
601\*[Gt]\*[Gt](0x3c.l)  string  PE\e0\e0  PE executable (MS-Windows)
602\*[Gt]\*[Gt](0x3c.l)  string  LX\e0\e0  LX executable (OS/2)
603.Ed
604.Pp
605This strategy of examining has a drawback: you must make sure that you
606eventually print something, or users may get empty output (such as when
607there is neither PE\e0\e0 nor LE\e0\e0 in the above example).
608.Pp
609If this indirect offset cannot be used directly, simple calculations are
610possible: appending
611.Em [+-*/%\*[Am]|^]number
612inside parentheses allows one to modify
613the value read from the file before it is used as an offset:
614.Bd -literal -offset indent
615# MS Windows executables are also valid MS-DOS executables
6160           string  MZ
617# sometimes, the value at 0x18 is less that 0x40 but there's still an
618# extended executable, simply appended to the file
619\*[Gt]0x18       leshort \*[Lt]0x40
620\*[Gt]\*[Gt](4.s*512) leshort 0x014c  COFF executable (MS-DOS, DJGPP)
621\*[Gt]\*[Gt](4.s*512) leshort !0x014c MZ executable (MS-DOS)
622.Ed
623.Pp
624Sometimes you do not know the exact offset as this depends on the length or
625position (when indirection was used before) of preceding fields.
626You can specify an offset relative to the end of the last up-level
627field using
628.Sq \*[Am]
629as a prefix to the offset:
630.Bd -literal -offset indent
6310           string  MZ
632\*[Gt]0x18       leshort \*[Gt]0x3f
633\*[Gt]\*[Gt](0x3c.l)  string  PE\e0\e0    PE executable (MS-Windows)
634# immediately following the PE signature is the CPU type
635\*[Gt]\*[Gt]\*[Gt]\*[Am]0       leshort 0x14c     for Intel 80386
636\*[Gt]\*[Gt]\*[Gt]\*[Am]0       leshort 0x184     for DEC Alpha
637.Ed
638.Pp
639Indirect and relative offsets can be combined:
640.Bd -literal -offset indent
6410             string  MZ
642\*[Gt]0x18         leshort \*[Lt]0x40
643\*[Gt]\*[Gt](4.s*512)   leshort !0x014c MZ executable (MS-DOS)
644# if it's not COFF, go back 512 bytes and add the offset taken
645# from byte 2/3, which is yet another way of finding the start
646# of the extended executable
647\*[Gt]\*[Gt]\*[Gt]\*[Am](2.s-514) string  LE      LE executable (MS Windows VxD driver)
648.Ed
649.Pp
650Or the other way around:
651.Bd -literal -offset indent
6520                 string  MZ
653\*[Gt]0x18             leshort \*[Gt]0x3f
654\*[Gt]\*[Gt](0x3c.l)        string  LE\e0\e0  LE executable (MS-Windows)
655# at offset 0x80 (-4, since relative offsets start at the end
656# of the up-level match) inside the LE header, we find the absolute
657# offset to the code area, where we look for a specific signature
658\*[Gt]\*[Gt]\*[Gt](\*[Am]0x7c.l+0x26) string  UPX     \eb, UPX compressed
659.Ed
660.Pp
661Or even both!
662.Bd -literal -offset indent
6630                string  MZ
664\*[Gt]0x18            leshort \*[Gt]0x3f
665\*[Gt]\*[Gt](0x3c.l)       string  LE\e0\e0 LE executable (MS-Windows)
666# at offset 0x58 inside the LE header, we find the relative offset
667# to a data area where we look for a specific signature
668\*[Gt]\*[Gt]\*[Gt]\*[Am](\*[Am]0x54.l-3)  string  UNACE  \eb, ACE self-extracting archive
669.Ed
670.Pp
671If you have to deal with offset/length pairs in your file, even the
672second value in a parenthesized expression can be taken from the file itself,
673using another set of parentheses.
674Note that this additional indirect offset is always relative to the
675start of the main indirect offset.
676.Bd -literal -offset indent
6770                 string       MZ
678\*[Gt]0x18             leshort      \*[Gt]0x3f
679\*[Gt]\*[Gt](0x3c.l)        string       PE\e0\e0 PE executable (MS-Windows)
680# search for the PE section called ".idata"...
681\*[Gt]\*[Gt]\*[Gt]\*[Am]0xf4          search/0x140 .idata
682# ...and go to the end of it, calculated from start+length;
683# these are located 14 and 10 bytes after the section name
684\*[Gt]\*[Gt]\*[Gt]\*[Gt](\*[Am]0xe.l+(-4)) string       PK\e3\e4 \eb, ZIP self-extracting archive
685.Ed
686.Pp
687If you have a list of known values at a particular continuation level,
688and you want to provide a switch-like default case:
689.Bd -literal -offset indent
690# clear that continuation level match
691\*[Gt]18	clear
692\*[Gt]18	lelong	1	one
693\*[Gt]18	lelong	2	two
694\*[Gt]18	default	x
695# print default match
696\*[Gt]\*[Gt]18	lelong	x	unmatched 0x%x
697.Ed
698.Sh SEE ALSO
699.Xr file __CSECTION__
700\- the command that reads this file.
701.Sh BUGS
702The formats
703.Dv long ,
704.Dv belong ,
705.Dv lelong ,
706.Dv melong ,
707.Dv short ,
708.Dv beshort ,
709and
710.Dv leshort
711do not depend on the length of the C data types
712.Dv short
713and
714.Dv long
715on the platform, even though the Single
716.Ux
717Specification implies that they do.  However, as OS X Mountain Lion has
718passed the Single
719.Ux
720Specification validation suite, and supplies a version of
721.Xr file __CSECTION__
722in which they do not depend on the sizes of the C data types and that is
723built for a 64-bit environment in which
724.Dv long
725is 8 bytes rather than 4 bytes, presumably the validation suite does not
726test whether, for example
727.Dv long
728refers to an item with the same size as the C data type
729.Dv long .
730There should probably be
731.Dv type
732names
733.Dv int8 ,
734.Dv uint8 ,
735.Dv int16 ,
736.Dv uint16 ,
737.Dv int32 ,
738.Dv uint32 ,
739.Dv int64 ,
740and
741.Dv uint64 ,
742and specified-byte-order variants of them,
743to make it clearer that those types have specified widths.
744.\"
745.\" From: guy@sun.uucp (Guy Harris)
746.\" Newsgroups: net.bugs.usg
747.\" Subject: /etc/magic's format isn't well documented
748.\" Message-ID: <2752@sun.uucp>
749.\" Date: 3 Sep 85 08:19:07 GMT
750.\" Organization: Sun Microsystems, Inc.
751.\" Lines: 136
752.\"
753.\" Here's a manual page for the format accepted by the "file" made by adding
754.\" the changes I posted to the S5R2 version.
755.\"
756.\" Modified for Ian Darwin's version of the file command.
757