1 (** Bitmatch library. *)
2 (* Copyright (C) 2008 Red Hat Inc., Richard W.M. Jones
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 * $Id: bitmatch.mli,v 1.18 2008-05-07 14:37:00 rjones Exp $
22 {{:#reference}Jump straight to the reference section for
23 documentation on types and functions}.
27 Bitmatch adds Erlang-style bitstrings and matching over bitstrings
28 as a syntax extension and library for OCaml. You can use
29 this module to both parse and generate binary formats, for
30 example, communications protocols, disk formats and binary files.
32 {{:http://et.redhat.com/~rjones/bitmatch/}OCaml bitmatch website}
36 A function which can parse IPv4 packets:
43 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
44 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
45 | 4 | IHL |Type of Service| Total Length |
46 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
47 | Identification |Flags| Fragment Offset |
48 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
49 | Time to Live | Protocol | Header Checksum |
50 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
52 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
53 | Destination Address |
54 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
56 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
58 | { 4 : 4; hdrlen : 4; tos : 8; length : 16;
59 identification : 16; flags : 3; fragoffset : 13;
60 ttl : 8; protocol : 8; checksum : 16;
63 options : (hdrlen-5)*32 : bitstring;
64 payload : -1 : bitstring } ->
67 printf " header length: %d * 32 bit words\n" hdrlen;
68 printf " type of service: %d\n" tos;
69 printf " packet length: %d bytes\n" length;
70 printf " identification: %d\n" identification;
71 printf " flags: %d\n" flags;
72 printf " fragment offset: %d\n" fragoffset;
73 printf " ttl: %d\n" ttl;
74 printf " protocol: %d\n" protocol;
75 printf " checksum: %d\n" checksum;
76 printf " source: %lx dest: %lx\n" source dest;
77 printf " header options + padding:\n";
78 Bitmatch.hexdump_bitstring stdout options;
79 printf " packet payload:\n";
80 Bitmatch.hexdump_bitstring stdout payload
83 eprintf "unknown IP version %d\n" version;
87 eprintf "data is smaller than one nibble:\n";
88 Bitmatch.hexdump_bitstring stderr pkt;
92 A program which can parse
93 {{:http://lxr.linux.no/linux/include/linux/ext3_fs.h}Linux EXT3 filesystem superblocks}:
96 let bits = Bitmatch.bitstring_of_file "tests/ext3_sb"
100 | { s_inodes_count : 32 : littleendian; (* Inodes count *)
101 s_blocks_count : 32 : littleendian; (* Blocks count *)
102 s_r_blocks_count : 32 : littleendian; (* Reserved blocks count *)
103 s_free_blocks_count : 32 : littleendian; (* Free blocks count *)
104 s_free_inodes_count : 32 : littleendian; (* Free inodes count *)
105 s_first_data_block : 32 : littleendian; (* First Data Block *)
106 s_log_block_size : 32 : littleendian; (* Block size *)
107 s_log_frag_size : 32 : littleendian; (* Fragment size *)
108 s_blocks_per_group : 32 : littleendian; (* # Blocks per group *)
109 s_frags_per_group : 32 : littleendian; (* # Fragments per group *)
110 s_inodes_per_group : 32 : littleendian; (* # Inodes per group *)
111 s_mtime : 32 : littleendian; (* Mount time *)
112 s_wtime : 32 : littleendian; (* Write time *)
113 s_mnt_count : 16 : littleendian; (* Mount count *)
114 s_max_mnt_count : 16 : littleendian; (* Maximal mount count *)
115 0xef53 : 16 : littleendian } -> (* Magic signature *)
117 printf "ext3 superblock:\n";
118 printf " s_inodes_count = %ld\n" s_inodes_count;
119 printf " s_blocks_count = %ld\n" s_blocks_count;
120 printf " s_free_inodes_count = %ld\n" s_free_inodes_count;
121 printf " s_free_blocks_count = %ld\n" s_free_blocks_count
124 eprintf "not an ext3 superblock!\n%!";
128 Constructing packets for a simple binary message
133 +---------------+---------------+--------------------------+
134 | type | subtype | parameter |
135 +---------------+---------------+--------------------------+
136 <-- 16 bits --> <-- 16 bits --> <------- 32 bits -------->
138 All fields are in network byte order.
141 let make_message typ subtype param =
149 {2 Loading, creating bitstrings}
151 The basic data type is the {!bitstring}, a string of bits of
152 arbitrary length. Bitstrings can be any length in bits and
153 operations do not need to be byte-aligned (although they will
154 generally be more efficient if they are byte-aligned).
156 Internally a bitstring is stored as a normal OCaml [string]
157 together with an offset and length, where the offset and length are
158 measured in bits. Thus one can efficiently form substrings of
159 bitstrings, overlay a bitstring on existing data, and load and save
160 bitstrings from files or other external sources.
162 To load a bitstring from a file use {!bitstring_of_file} or
163 {!bitstring_of_chan}.
165 There are also functions to create bitstrings from arbitrary data.
166 See the {{:#reference}reference} below.
168 {2 Matching bitstrings with patterns}
170 Use the [bitmatch] operator (part of the syntax extension) to break
171 apart a bitstring into its fields. [bitmatch] works a lot like the
172 OCaml [match] operator.
174 The general form of [bitmatch] is:
176 [bitmatch {] {i bitstring-expression} [} with]
178 [| {] {i pattern} [} ->] {i code}
180 [| {] {i pattern} [} ->] {i code}
184 As with normal match, the statement attempts to match the
185 bitstring against each pattern in turn. If none of the patterns
186 match then the standard library [Match_failure] exception is
189 Patterns look a bit different from normal match patterns. The
190 consist of a list of bitfields separated by [;] where each bitfield
191 contains a bind variable, the width (in bits) of the field, and
192 other information. Some example patterns:
197 | { version : 8; name : 8; param : 8 } -> ...
199 (* Bitstring of at least 3 bytes. First byte is the version
200 number, second byte is a field called name, third byte is
201 a field called parameter. *)
204 printf "flag is %b\n" flag
206 (* A single flag bit (mapped into an OCaml boolean). *)
208 | { len : 4; data : 1+len } ->
209 printf "len = %d, data = 0x%Lx\n" len data
211 (* A 4-bit length, followed by 1-16 bits of data, where the
212 length of the data is computed from len. *)
214 | { ipv6_source : 128 : bitstring;
215 ipv6_dest : 128 : bitstring } -> ...
217 (* IPv6 source and destination addresses. Each is 128 bits
218 and is mapped into a bitstring type which will be a substring
219 of the main bitstring expression. *)
222 You can also add conditional when-clauses:
226 when version = 4 || version = 6 -> ...
228 (* Only match and run the code when version is 4 or 6. If
229 it isn't we will drop through to the next case. *)
232 Note that the pattern is only compared against the first part of
233 the bitstring (there may be more data in the bitstring following
234 the pattern, which is not matched). In terms of regular
235 expressions you might say that the pattern matches [^pattern], not
236 [^pattern$]. To ensure that the bitstring contains only the
237 pattern, add a length -1 bitstring to the end and test that its
238 length is zero in the when-clause:
242 rest : -1 : bitstring }
243 when Bitmatch.bitstring_length rest = 0 -> ...
245 (* Only matches exactly 4 bits. *)
248 Normally the first part of each field is a binding variable,
249 but you can also match a constant, as in:
252 | { (4|6) : 4 } -> ...
254 (* Only matches if the first 4 bits contain either
255 the integer 4 or the integer 6. *)
258 One may also match on strings:
261 | { "MAGIC" : 5*8 : string } -> ...
263 (* Only matches if the string "MAGIC" appears at the start
267 {3:patternfieldreference Pattern field reference}
269 The exact format of each pattern field is:
271 [pattern : length [: qualifier [,qualifier ...]]]
273 [pattern] is the pattern, binding variable name, or constant to
274 match. [length] is the length in bits which may be either a
275 constant or an expression. The length expression is just an OCaml
276 expression and can use any values defined in the program, and refer
277 back to earlier fields (but not to later fields).
279 Integers can only have lengths in the range \[1..64\] bits. See the
280 {{:#integertypes}integer types} section below for how these are
281 mapped to the OCaml int/int32/int64 types. This is checked
282 at compile time if the length expression is constant, otherwise it is
283 checked at runtime and you will get a runtime exception eg. in
284 the case of a computed length expression.
286 A bitstring field of length -1 matches all the rest of the
287 bitstring (thus this is only useful as the last field in a
290 A bitstring field of length 0 matches an empty bitstring
291 (occasionally useful when matching optional subfields).
293 Qualifiers are a list of identifiers which control the type,
294 signedness and endianness of the field. Permissible qualifiers are:
296 - [int] (field has an integer type)
297 - [string] (field is a string type)
298 - [bitstring] (field is a bitstring type)
299 - [signed] (field is signed)
300 - [unsigned] (field is unsigned)
301 - [bigendian] (field is big endian - a.k.a network byte order)
302 - [littleendian] (field is little endian - a.k.a Intel byte order)
303 - [nativeendian] (field is same endianness as the machine)
305 The default settings are [int], [unsigned], [bigendian].
307 Note that many of these qualifiers cannot be used together,
308 eg. bitstrings do not have endianness. The syntax extension should
309 give you a compile-time error if you use incompatible qualifiers.
311 {3 Other cases in bitmatch}
313 As well as a list of fields, it is possible to name the
314 bitstring and/or have a default match case:
319 (* Default match case. *)
321 | { _ } as pkt -> ...
323 (* Default match case, with 'pkt' bound to the whole bitstring. *)
326 {2 Constructing bitstrings}
328 Bitstrings may be constructed using the [BITSTRING] operator (as an
329 expression). The [BITSTRING] operator takes a list of fields,
330 similar to the list of fields for matching:
341 (* Constructs a 16-bit bitstring with the first four bits containing
342 the integer 1, and the following 12 bits containing the integer 10,
343 arranged in network byte order. *)
345 Bitmatch.hexdump_bitstring stdout bits ;;
353 The format of each field is the same as for pattern fields (see
354 {{:#patternfieldreference}Pattern field reference section}), and
355 things like computed length fields, fixed value fields, insertion
356 of bitstrings within bitstrings, etc. are all supported.
358 {3 Construction exception}
360 The [BITSTRING] operator may throw a {!Construct_failure}
361 exception at runtime.
363 Runtime errors include:
365 - int field length not in the range \[1..64\]
366 - a bitstring with a length declared which doesn't have the
367 same length at runtime
368 - trying to insert an out of range value into an int field
369 (eg. an unsigned int field which is 2 bits wide can only
370 take values in the range \[0..3\]).
372 {2:integertypes Integer types}
374 Integer types are mapped to OCaml types [bool], [int], [int32] or
375 [int64] using a system which tries to ensure that (a) the types are
376 reasonably predictable and (b) the most efficient type is
379 The rules are slightly different depending on whether the bit
380 length expression in the field is a compile-time constant or a
383 Detection of compile-time constants is quite simplistic so only an
384 immediate, simple integer is recognised as a constant and anything
385 else is considered a computed expression, even expressions such as
386 [5-2] which are obviously (to our eyes) constant.
388 In any case the bit size of an integer is limited to the range
389 \[1..64\]. This is detected as a compile-time error if that is
390 possible, otherwise a runtime check is added which can throw an
391 [Invalid_argument] exception.
396 Bit size ---- OCaml type ----
397 Constant Computed expression
405 A possible future extension may allow people with 64 bit computers
406 to specify a more optimal [int] type for bit sizes in the range
407 [32..63]. If this was implemented then such code {i could not even
408 be compiled} on 32 bit platforms, so it would limit portability.
410 Another future extension may be to allow computed
411 expressions to assert min/max range for the bit size,
412 allowing a more efficient data type than int64 to be
413 used. (Of course under such circumstances there would
414 still need to be a runtime check to enforce the
419 Using the compiler directly you can do:
422 ocamlc -I +bitmatch \
423 -pp "camlp4o `ocamlc -where`/bitmatch/pa_bitmatch.cmo" \
424 bitmatch.cma test.ml -o test
427 Simpler method using findlib:
431 -package bitmatch.syntax -syntax bitmatch.syntax \
432 -linkpkg test.ml -o test
435 {2 Security and type safety}
437 {3 Security on input}
439 The main concerns for input are buffer overflows and denial
442 It is believed that this library is robust against attempted buffer
443 overflows. In addition to OCaml's normal bounds checks, we check
444 that field lengths are >= 0, and many additional checks.
446 Denial of service attacks are more problematic. We only work
447 forwards through the bitstring, thus computation will eventually
448 terminate. As for computed lengths, code such as this is thought
454 buffer : Int64.to_int len : bitstring } ->
457 The [len] field can be set arbitrarily large by an attacker, but
458 when pattern-matching against the [buffer] field this merely causes
459 a test such as [if len <= remaining_size] to fail. Even if the
460 length is chosen so that [buffer] bitstring is allocated, the
461 allocation of sub-bitstrings is efficient and doesn't involve an
462 arbitary-sized allocation or any copying.
464 However the above does not necessarily apply to strings used in
465 matching, since they may cause the library to use the
466 {!Bitmatch.string_of_bitstring} function, which allocates a string.
467 So you should take care if you use the [string] type particularly
468 with a computed length that is derived from external input.
470 The main protection against attackers should be to ensure that the
471 main program will only read input bitstrings up to a certain
472 length, which is outside the scope of this library.
474 {3 Security on output}
476 As with the input side, computed lengths are believed to be
480 let len = read_untrusted_source () in
481 let buffer = allocate_bitstring () in
483 buffer : len : bitstring
487 This code merely causes a check that buffer's length is the same as
488 [len]. However the program function [allocate_bitstring] must
489 refuse to allocate an oversized buffer (but that is outside the
490 scope of this library).
492 {3 Order of evaluation}
494 In [bitmatch] statements, fields are evaluated left to right.
496 Note that the when-clause is evaluated {i last}, so if you are
497 relying on the when-clause to filter cases then your code may do a
498 lot of extra and unncessary pattern-matching work on fields which
499 may never be needed just to evaluate the when-clause. You can
500 usually rearrange the code to do only the first part of the match,
501 followed by the when-clause, followed by a second inner bitmatch.
505 The current implementation is believed to be fully type-safe,
506 and makes compile and run-time checks where appropriate. If
507 you find a case where a check is missing please submit a
508 bug report or a patch.
512 These are thought to be the current limits:
514 Integers: \[1..64\] bits.
516 Bitstrings (32 bit platforms): maximum length is limited
517 by the string size, ie. 16 MBytes.
519 Bitstrings (64 bit platforms): maximum length is thought to be
520 limited by the string size, ie. effectively unlimited.
522 Bitstrings must be loaded into memory before we can match against
523 them. Thus available memory may be considered a limit for some
526 {2:reference Reference}
530 type bitstring = string * int * int
531 (** [bitstring] is the basic type used to store bitstrings.
533 The type contains the underlying data (a string),
534 the current bit offset within the string and the
535 current bit length of the string (counting from the
536 bit offset). Note that the offset and length are
537 in {b bits}, not bytes.
539 Normally you don't need to use the bitstring type
540 directly, since there are functions and syntax
541 extensions which hide the details.
543 See also {!bitstring_of_string}, {!bitstring_of_file},
544 {!hexdump_bitstring}, {!bitstring_length}.
547 (** {3 Exceptions} *)
549 exception Construct_failure of string * string * int * int
550 (** [Construct_failure (message, file, line, char)] may be
551 raised by the [BITSTRING] constructor.
553 Common reasons are that values are out of range of
554 the fields that contain them, or that computed lengths
555 are impossible (eg. negative length bitfields).
557 [message] is the error message.
559 [file], [line] and [char] point to the original source
560 location of the [BITSTRING] constructor that failed.
563 (** {3 Bitstrings} *)
565 val empty_bitstring : bitstring
566 (** [empty_bitstring] is the empty, zero-length bitstring. *)
568 val create_bitstring : int -> bitstring
569 (** [create_bitstring n] creates an [n] bit bitstring
570 containing all zeroes. *)
572 val make_bitstring : int -> char -> bitstring
573 (** [make_bitstring n c] creates an [n] bit bitstring
574 containing the repeated 8 bit pattern in [c].
576 For example, [make_bitstring 16 '\x5a'] will create
577 the bitstring [0x5a5a] or in binary [0101 1010 0101 1010].
579 Note that the length is in bits, not bytes. *)
581 val bitstring_of_string : string -> bitstring
582 (** [bitstring_of_string str] creates a bitstring
583 of length [String.length str * 8] (bits) containing the
586 Note that the bitstring uses [str] as the underlying
587 string (see the representation of {!bitstring}) so you
588 should not change [str] after calling this. *)
590 val bitstring_of_file : string -> bitstring
591 (** [bitstring_of_file filename] loads the named file
594 val bitstring_of_chan : in_channel -> bitstring
595 (** [bitstring_of_chan chan] loads the contents of
596 the input channel [chan] as a bitstring.
598 The length of the final bitstring is determined
599 by the remaining input in [chan], but will always
600 be a multiple of 8 bits.
602 See also {!bitstring_of_chan_max}. *)
604 val bitstring_of_chan_max : in_channel -> int -> bitstring
605 (** [bitstring_of_chan_max chan max] works like
606 {!bitstring_of_chan} but will only read up to
607 [max] bytes from the channel (or fewer if the end of input
608 occurs before that). *)
610 val bitstring_of_file_descr : Unix.file_descr -> bitstring
611 (** [bitstring_of_file_descr fd] loads the contents of
612 the file descriptor [fd] as a bitstring.
614 See also {!bitstring_of_chan}, {!bitstring_of_file_descr_max}. *)
616 val bitstring_of_file_descr_max : Unix.file_descr -> int -> bitstring
617 (** [bitstring_of_file_descr_max fd max] works like
618 {!bitstring_of_file_descr} but will only read up to
619 [max] bytes from the channel (or fewer if the end of input
620 occurs before that). *)
622 val bitstring_length : bitstring -> int
623 (** [bitstring_length bitstring] returns the length of
624 the bitstring in bits. *)
626 val string_of_bitstring : bitstring -> string
627 (** [string_of_bitstring bitstring] converts a bitstring to a string
628 (eg. to allow comparison).
630 This function is inefficient. In the best case when the bitstring
631 is nicely byte-aligned we do a [String.sub] operation. If the
632 bitstring isn't aligned then this involves a lot of bit twiddling
633 and is particularly inefficient. *)
635 (** {3 Printing bitstrings} *)
637 val hexdump_bitstring : out_channel -> bitstring -> unit
638 (** [hexdump_bitstring chan bitstring] prints the bitstring
639 to the output channel in a format similar to the
640 Unix command [hexdump -C]. *)
642 (** {3 Bitstring buffer} *)
646 val create : unit -> t
647 val contents : t -> bitstring
648 val add_bits : t -> string -> int -> unit
649 val add_bit : t -> bool -> unit
650 val add_byte : t -> int -> unit
652 (** Buffers are mainly used by the [BITSTRING] constructor, but
653 may also be useful for end users. They work much like the
654 standard library [Buffer] module. *)
656 (** {3 Miscellaneous} *)
659 (** Set this variable to true to enable extended debugging.
660 This only works if debugging was also enabled in the
661 [pa_bitmatch.ml] file at compile time, otherwise it
666 (* Private functions, called from generated code. Do not use
667 * these directly - they are not safe.
670 val extract_bitstring : string -> int -> int -> int -> bitstring * int * int
672 val extract_remainder : string -> int -> int -> bitstring * int * int
674 val extract_bit : string -> int -> int -> int -> bool * int * int
676 val extract_char_unsigned : string -> int -> int -> int -> int * int * int
678 val extract_int_be_unsigned : string -> int -> int -> int -> int * int * int
680 val extract_int_le_unsigned : string -> int -> int -> int -> int * int * int
682 val extract_int32_be_unsigned : string -> int -> int -> int -> int32 * int * int
684 val extract_int32_le_unsigned : string -> int -> int -> int -> int32 * int * int
686 val extract_int64_be_unsigned : string -> int -> int -> int -> int64 * int * int
688 val construct_bit : Buffer.t -> bool -> int -> unit
690 val construct_char_unsigned : Buffer.t -> int -> int -> exn -> unit
692 val construct_int_be_unsigned : Buffer.t -> int -> int -> exn -> unit
694 val construct_int64_be_unsigned : Buffer.t -> int64 -> int -> exn -> unit
696 val construct_string : Buffer.t -> string -> unit