--- CBOR-XS/README 2014/01/05 14:24:54 1.13 +++ CBOR-XS/README 2020/12/08 08:29:44 1.21 @@ -35,6 +35,9 @@ compress the data later and speed is less important you might want to compare both formats first). + The primary goal of this module is to be *correct* and the secondary + goal is to be *fast*. To reach the latter goal it was written in C. + To give you a general idea about speed, with texts in the megabyte range, "CBOR::XS" usually encodes roughly twice as fast as Storable or JSON::XS and decodes about 15%-30% faster than those. The shorter the @@ -49,9 +52,6 @@ "allow_sharing" and "allow_cycles"), string deduplication (see "pack_strings") and scalar references (always enabled). - The primary goal of this module is to be *correct* and the secondary - goal is to be *fast*. To reach the latter goal it was written in C. - See MAPPING, below, on how CBOR::XS maps perl values to CBOR values and vice versa. @@ -81,6 +81,20 @@ my $cbor = CBOR::XS->new->encode ({a => [1,2]}); + $cbor = new_safe CBOR::XS + Create a new, safe/secure CBOR::XS object. This is similar to "new", + but configures the coder object to be safe to use with untrusted + data. Currently, this is equivalent to: + + my $cbor = CBOR::XS + ->new + ->forbid_objects + ->filter (\&CBOR::XS::safe_filter) + ->max_size (1e8); + + But is more future proof (it is better to crash because of a change + than to be exploited in other ways). + $cbor = $cbor->max_depth ([$maximum_nesting_depth]) $max_depth = $cbor->get_max_depth Sets the maximum nesting level (default 512) accepted while encoding @@ -103,7 +117,7 @@ value has been chosen to be as large as typical operating systems allow without crashing. - See SECURITY CONSIDERATIONS, below, for more info on why this is + See "SECURITY CONSIDERATIONS", below, for more info on why this is useful. $cbor = $cbor->max_size ([$maximum_string_size]) @@ -117,7 +131,7 @@ If no argument is given, the limit check will be deactivated (same as when 0 is specified). - See SECURITY CONSIDERATIONS, below, for more info on why this is + See "SECURITY CONSIDERATIONS", below, for more info on why this is useful. $cbor = $cbor->allow_unknown ([$enable]) @@ -143,7 +157,7 @@ This means that such values will only be encoded once, and will not result in a deep cloning of the value on decode, in decoders supporting the value sharing extension. This also makes it possible - to encode cyclic data structures (which need "allow_cycles" to ne + to encode cyclic data structures (which need "allow_cycles" to be enabled to be decoded by this module). It is recommended to leave it off unless you know your communication @@ -154,7 +168,7 @@ Detecting shared values incurs a runtime overhead when values are encoded that have a reference counter large than one, and might unnecessarily increase the encoded size, as potentially shared - values are encode as shareable whether or not they are actually + values are encoded as shareable whether or not they are actually shared. At the moment, only targets of references can be shared (e.g. @@ -180,8 +194,32 @@ If $enable is false (the default), then "decode" will throw an error when it encounters a self-referential/cyclic data structure. + FUTURE DIRECTION: the motivation behind this option is to avoid + *real* cycles - future versions of this module might chose to decode + cyclic data structures using weak references when this option is + off, instead of throwing an error. + This option does not affect "encode" in any way - shared values and - references will always be decoded properly if present. + references will always be encoded properly if present. + + $cbor = $cbor->forbid_objects ([$enable]) + $enabled = $cbor->get_forbid_objects + Disables the use of the object serialiser protocol. + + If $enable is true (or missing), then "encode" will will throw an + exception when it encounters perl objects that would be encoded + using the perl-object tag (26). When "decode" encounters such tags, + it will fall back to the general filter/tagged logic as if this were + an unknown tag (by default resulting in a "CBOR::XC::Tagged" + object). + + If $enable is false (the default), then "encode" will use the + Types::Serialiser object serialisation protocol to serialise objects + into perl-object tags, and "decode" will do the same to decode such + tags. + + See "SECURITY CONSIDERATIONS", below, for more info on why + forbidding this protocol can be useful. $cbor = $cbor->pack_strings ([$enable]) $enabled = $cbor->get_pack_strings @@ -202,6 +240,51 @@ This option does not affect "decode" in any way - string references will always be decoded properly if present. + $cbor = $cbor->text_keys ([$enable]) + $enabled = $cbor->get_text_keys + If $enabled is true (or missing), then "encode" will encode all perl + hash keys as CBOR text strings/UTF-8 string, upgrading them as + needed. + + If $enable is false (the default), then "encode" will encode hash + keys normally - upgraded perl strings (strings internally encoded as + UTF-8) as CBOR text strings, and downgraded perl strings as CBOR + byte strings. + + This option does not affect "decode" in any way. + + This option is useful for interoperability with CBOR decoders that + don't treat byte strings as a form of text. It is especially useful + as Perl gives very little control over hash keys. + + Enabling this option can be slow, as all downgraded hash keys that + are encoded need to be scanned and converted to UTF-8. + + $cbor = $cbor->text_strings ([$enable]) + $enabled = $cbor->get_text_strings + This option works similar to "text_keys", above, but works on all + strings (including hash keys), so "text_keys" has no further effect + after enabling "text_strings". + + If $enabled is true (or missing), then "encode" will encode all perl + strings as CBOR text strings/UTF-8 strings, upgrading them as + needed. + + If $enable is false (the default), then "encode" will encode strings + normally (but see "text_keys") - upgraded perl strings (strings + internally encoded as UTF-8) as CBOR text strings, and downgraded + perl strings as CBOR byte strings. + + This option does not affect "decode" in any way. + + This option has similar advantages and disadvantages as "text_keys". + In addition, this option effectively removes the ability to + automatically encode byte strings, which might break some "FREEZE" + and "TO_CBOR" methods that rely on this. + + A workaround is to use explicit type casts, which are unaffected by + this option. + $cbor = $cbor->validate_utf8 ([$enable]) $enabled = $cbor->get_validate_utf8 If $enable is true (or missing), then "decode" will validate that @@ -214,7 +297,7 @@ If $enable is false (the default), then "decode" will blindly accept UTF-8 data, marking them as valid UTF-8 in the resulting data - structure regardless of whether thats true or not. + structure regardless of whether that's true or not. Perl isn't too happy about corrupted UTF-8 in strings, but should generally not crash or do similarly evil things. Extensions might be @@ -251,7 +334,20 @@ looks up the tag in the %CBOR::XS::FILTER hash. If an entry exists it must be a code reference that is called with tag and value, and is responsible for decoding the value. If no entry exists, it - returns no values. + returns no values. "CBOR::XS" provides a number of default filter + functions already, the the %CBOR::XS::FILTER hash can be freely + extended with more. + + "CBOR::XS" additionally provides an alternative filter function that + is supposed to be safe to use with untrusted data (which the default + filter might not), called "CBOR::XS::safe_filter", which works the + same as the "default_filter" but uses the %CBOR::XS::SAFE_FILTER + variable instead. It is prepopulated with the tag decoding functions + that are deemed safe (basically the same as %CBOR::XS::FILTER + without all the bignum tags), and can be extended by user code as + wlel, although, obviously, one should be very careful about adding + decoding functions here, since the expectation is that they are safe + to use on untrusted data, after all. Example: decode all tags not handled internally into "CBOR::XS::Tagged" objects, with no other special handling (useful @@ -268,6 +364,26 @@ "tag 1347375694 value $value" }; + Example: provide your own filter function that looks up tags in your + own hash: + + my %my_filter = ( + 998347484 => sub { + my ($tag, $value); + + "tag 998347484 value $value" + }; + ); + + my $coder = CBOR::XS->new->filter (sub { + &{ $my_filter{$_[0]} or return } + }); + + Example: use the safe filter function (see "SECURITY CONSIDERATIONS" + for more considerations on security). + + CBOR::XS->new->filter (\&CBOR::XS::safe_filter)->decode ($cbor_data); + $cbor_data = $cbor->encode ($perl_scalar) Converts the given Perl data structure (a scalar value) to its CBOR representation. @@ -284,7 +400,9 @@ This is useful if your CBOR texts are not delimited by an outer protocol and you need to know where the first CBOR string ends amd - the next one starts. + the next one starts - CBOR strings are self-delimited, so it is + possible to concatenate CBOR strings without any delimiters or size + fields and recover their data. CBOR::XS->new->decode_prefix ("......") => ("...", 3) @@ -343,9 +461,9 @@ that subsequent calls to "incr_parse" or "incr_parse_multiple" start to parse a new CBOR value from the beginning of the $buffer again. - This method can be caled at any time, but it *must* be called if you - want to change your $buffer or there was a decoding error and you - want to reuse the $cbor object for future incremental parsings. + This method can be called at any time, but it *must* be called if + you want to change your $buffer or there was a decoding error and + you want to reuse the $cbor object for future incremental parsings. MAPPING This section describes how CBOR::XS maps Perl values to CBOR values and @@ -406,7 +524,7 @@ Perl hash references become CBOR maps. As there is no inherent ordering in hash keys (or CBOR maps), they will usually be encoded in a pseudo-random order. This order can be different each time a - hahs is encoded. + hash is encoded. Currently, tied hashes will use the indefinite-length format, while normal hashes will use the fixed-length format. @@ -465,15 +583,21 @@ $x .= ""; # another, more awkward way to stringify print $x; # perl does it for you, too, quite often - You can force whether a string ie encoded as byte or text string by - using "utf8::upgrade" and "utf8::downgrade"): + You can force whether a string is encoded as byte or text string by + using "utf8::upgrade" and "utf8::downgrade" (if "text_strings" is + disabled). utf8::upgrade $x; # encode $x as text string utf8::downgrade $x; # encode $x as byte string + More options are available, see "TYPE CASTS", below, and the + "text_keys" and "text_strings" options. + Perl doesn't define what operations up- and downgrade strings, so if the difference between byte and text is important, you should up- or - downgrade your string as late as possible before encoding. + downgrade your string as late as possible before encoding. You can + also force the use of CBOR text strings by using "text_keys" or + "text_strings". You can force the type to be a CBOR number by numifying it: @@ -492,6 +616,69 @@ than IEEE double to represent numerical values are supported, but might suffer loss of precision. + TYPE CASTS + EXPERIMENTAL: As an experimental extension, "CBOR::XS" allows you to + force specific cbor types to be used when encoding. That allows you to + encode types not normally accessible (e.g. half floats) as well as force + string types even when "text_strings" is in effect. + + Type forcing is done by calling a special "cast" function which keeps a + copy of the value and returns a new value that can be handed over to any + CBOR encoder function. + + The following casts are currently available (all of which are unary + operators): + + CBOR::XS::as_int $value + Forces the value to be encoded as some form of (basic, not bignum) + integer type. + + CBOR::XS::as_text $value + Forces the value to be encoded as (UTF-8) text values. + + CBOR::XS::as_bytes $value + Forces the value to be encoded as a (binary) string value. + + Example: encode a perl string as binary even though "text_strings" + is in effect. + + CBOR::XS->new->text_strings->encode ([4, "text", CBOR::XS::bytes "bytevalue"]); + + CBOR::XS::as_bool $value + Converts a Perl boolean (which can be any kind of scalar) into a + CBOR boolean. Strictly the same, but shorter to write, than: + + $value ? Types::Serialiser::true : Types::Serialiser::false + + CBOR::XS::as_float16 $value + Forces half-float (IEEE 754 binary16) encoding of the given value. + + CBOR::XS::as_float32 $value + Forces single-float (IEEE 754 binary32) encoding of the given value. + + CBOR::XS::as_float64 $value + Forces double-float (IEEE 754 binary64) encoding of the given value. + + CBOR::XS::as_cbor $cbor_text + Not a type cast per-se, this type cast forces the argument to eb + encoded as-is. This can be used to embed pre-encoded CBOR data. + + Note that no checking on the validity of the $cbor_text is done - + it's the callers responsibility to correctly encode values. + + CBOR::XS::as_map [key => value...] + Treat the array reference as key value pairs and output a CBOR map. + This allows you to generate CBOR maps with arbitrary key types (or, + if you don't care about semantics, duplicate keys or prairs in a + custom order), which is otherwise hard to do with Perl. + + The single argument must be an array reference with an even number + of elements. + + Example: encode a CBOR map with a string and an integer as keys. + + encode_cbor CBOR::XS::as_map [string => "value", 5 => "value"] + OBJECT SERIALISATION This module implements both a CBOR-specific and the generic Types::Serialier object serialisation protocol. The following @@ -578,7 +765,6 @@ sub URI::THAW { my ($class, $serialiser, $uri) = @_; - $class->new ($uri) } @@ -684,14 +870,14 @@ ENFORCED TAGS These tags are always handled when decoding, and their handling cannot - be overriden by the user. + be overridden by the user. 26 (perl-object, ) These tags are automatically created (and decoded) for serialisable objects using the "FREEZE/THAW" methods (the Types::Serialier object serialisation protocol). See "OBJECT SERIALISATION" for details. - 28, 29 (shareable, sharedref, L ) + 28, 29 (shareable, sharedref, ) These tags are automatically decoded when encountered (and they do not result in a cyclic data structure, see "allow_cycles"), resulting in shared values in the decoded object. They are only @@ -710,15 +896,15 @@ considered too unimportant to be supported in the encoder. The decoder, however, will decode these values as shared values. - 256, 25 (stringref-namespace, stringref, L + 256, 25 (stringref-namespace, stringref, ) These tags are automatically decoded when encountered. They are only encoded, however, when "pack_strings" is enabled. 22098 (indirection, ) This tag is automatically generated when a reference are encountered - (with the exception of hash and array refernces). It is converted to - a reference when decoding. + (with the exception of hash and array references). It is converted + to a reference when decoding. 55799 (self-describe CBOR, RFC 7049) This value is not generated on encoding (unless explicitly requested @@ -726,8 +912,8 @@ NON-ENFORCED TAGS These tags have default filters provided when decoding. Their handling - can be overriden by changing the %CBOR::XS::FILTER entry for the tag, or - by providing a custom "filter" callback when decoding. + can be overridden by changing the %CBOR::XS::FILTER entry for the tag, + or by providing a custom "filter" callback when decoding. When they result in decoding into a specific Perl class, the module usually provides a corresponding "TO_CBOR" method as well. @@ -752,15 +938,23 @@ "Math::BigInt::TO_CBOR" method encodes "small" bigints into normal CBOR integers, and others into positive/negative CBOR bignums. - 4, 5 (decimal fraction/bigfloat) + 4, 5, 264, 265 (decimal fraction/bigfloat) Both decimal fractions and bigfloats are decoded into Math::BigFloat objects. The corresponding "Math::BigFloat::TO_CBOR" method *always* - encodes into a decimal fraction. + encodes into a decimal fraction (either tag 4 or 264). + + NaN and infinities are not encoded properly, as they cannot be + represented in CBOR. + + See "BIGNUM SECURITY CONSIDERATIONS" for more info. - CBOR cannot represent bigfloats with *very* large exponents - - conversion of such big float objects is undefined. + 30 (rational numbers) + These tags are decoded into Math::BigRat objects. The corresponding + "Math::BigRat::TO_CBOR" method encodes rational numbers with + denominator 1 via their numerator only, i.e., they become normal + integers or "bignums". - Also, NaN and infinities are not encoded properly. + See "BIGNUM SECURITY CONSIDERATIONS" for more info. 21, 22, 23 (expected later JSON conversion) CBOR::XS is not a CBOR-to-JSON converter, and will simply ignore @@ -784,38 +978,140 @@ CBOR intact. SECURITY CONSIDERATIONS - When you are using CBOR in a protocol, talking to untrusted potentially - hostile creatures requires relatively few measures. + Tl;dr... if you want to decode or encode CBOR from untrusted sources, + you should start with a coder object created via "new_safe" (which + implements the mitigations explained below): + + my $coder = CBOR::XS->new_safe; + + my $data = $coder->decode ($cbor_text); + my $cbor = $coder->encode ($data); + + Longer version: When you are using CBOR in a protocol, talking to + untrusted potentially hostile creatures requires some thought: + + Security of the CBOR decoder itself + First and foremost, your CBOR decoder should be secure, that is, + should not have any buffer overflows or similar bugs that could + potentially be exploited. Obviously, this module should ensure that + and I am trying hard on making that true, but you never know. + + CBOR::XS can invoke almost arbitrary callbacks during decoding + CBOR::XS supports object serialisation - decoding CBOR can cause + calls to *any* "THAW" method in *any* package that exists in your + process (that is, CBOR::XS will not try to load modules, but any + existing "THAW" method or function can be called, so they all have + to be secure). + + Less obviously, it will also invoke "TO_CBOR" and "FREEZE" methods - + even if all your "THAW" methods are secure, encoding data structures + from untrusted sources can invoke those and trigger bugs in those. + + So, if you are not sure about the security of all the modules you + have loaded (you shouldn't), you should disable this part using + "forbid_objects" or using "new_safe". + + CBOR can be extended with tags that call library code + CBOR can be extended with tags, and "CBOR::XS" has a registry of + conversion functions for many existing tags that can be extended via + third-party modules (see the "filter" method). + + If you don't trust these, you should configure the "safe" filter + function, "CBOR::XS::safe_filter" ("new_safe" does this), which by + default only includes conversion functions that are considered + "safe" by the author (but again, they can be extended by third party + modules). + + Depending on your level of paranoia, you can use the "safe" filter: + + $cbor->filter (\&CBOR::XS::safe_filter); + + ... your own filter... + + $cbor->filter (sub { ... do your stuffs here ... }); + + ... or even no filter at all, disabling all tag decoding: + + $cbor->filter (sub { }); + + This is never a problem for encoding, as the tag mechanism only + exists in CBOR texts. + + Resource-starving attacks: object memory usage + You need to avoid resource-starving attacks. That means you should + limit the size of CBOR data you accept, or make sure then when your + resources run out, that's just fine (e.g. by using a separate + process that can crash safely). The size of a CBOR string in octets + is usually a good indication of the size of the resources required + to decode it into a Perl structure. While CBOR::XS can check the + size of the CBOR text (using "max_size" - done by "new_safe"), it + might be too late when you already have it in memory, so you might + want to check the size before you accept the string. + + As for encoding, it is possible to construct data structures that + are relatively small but result in large CBOR texts (for example by + having an array full of references to the same big data structure, + which will all be deep-cloned during encoding by default). This is + rarely an actual issue (and the worst case is still just running out + of memory), but you can reduce this risk by using "allow_sharing". + + Resource-starving attacks: stack overflows + CBOR::XS recurses using the C stack when decoding objects and + arrays. The C stack is a limited resource: for instance, on my amd64 + machine with 8MB of stack size I can decode around 180k nested + arrays but only 14k nested CBOR objects (due to perl itself + recursing deeply on croak to free the temporary). If that is + exceeded, the program crashes. To be conservative, the default + nesting limit is set to 512. If your process has a smaller stack, + you should adjust this setting accordingly with the "max_depth" + method. + + Resource-starving attacks: CPU en-/decoding complexity + CBOR::XS will use the Math::BigInt, Math::BigFloat and Math::BigRat + libraries to represent encode/decode bignums. These can be very slow + (as in, centuries of CPU time) and can even crash your program (and + are generally not very trustworthy). See the next section on bignum + security for details. + + Data breaches: leaking information in error messages + CBOR::XS might leak contents of your Perl data structures in its + error messages, so when you serialise sensitive information you + might want to make sure that exceptions thrown by CBOR::XS will not + end up in front of untrusted eyes. + + Something else... + Something else could bomb you, too, that I forgot to think of. In + that case, you get to keep the pieces. I am always open for hints, + though... + +BIGNUM SECURITY CONSIDERATIONS + CBOR::XS provides a "TO_CBOR" method for both Math::BigInt and + Math::BigFloat that tries to encode the number in the simplest possible + way, that is, either a CBOR integer, a CBOR bigint/decimal fraction (tag + 4) or an arbitrary-exponent decimal fraction (tag 264). Rational numbers + (Math::BigRat, tag 30) can also contain bignums as members. + + CBOR::XS will also understand base-2 bigfloat or arbitrary-exponent + bigfloats (tags 5 and 265), but it will never generate these on its own. + + Using the built-in Math::BigInt::Calc support, encoding and decoding + decimal fractions is generally fast. Decoding bigints can be slow for + very big numbers (tens of thousands of digits, something that could + potentially be caught by limiting the size of CBOR texts), and decoding + bigfloats or arbitrary-exponent bigfloats can be *extremely* slow + (minutes, decades) for large exponents (roughly 40 bit and longer). + + Additionally, Math::BigInt can take advantage of other bignum libraries, + such as Math::GMP, which cannot handle big floats with large exponents, + and might simply abort or crash your program, due to their code quality. + + This can be a concern if you want to parse untrusted CBOR. If it is, you + might want to disable decoding of tag 2 (bigint) and 3 (negative bigint) + types. You should also disable types 5 and 265, as these can be slow + even without bigints. - First of all, your CBOR decoder should be secure, that is, should not - have any buffer overflows. Obviously, this module should ensure that and - I am trying hard on making that true, but you never know. - - Second, you need to avoid resource-starving attacks. That means you - should limit the size of CBOR data you accept, or make sure then when - your resources run out, that's just fine (e.g. by using a separate - process that can crash safely). The size of a CBOR string in octets is - usually a good indication of the size of the resources required to - decode it into a Perl structure. While CBOR::XS can check the size of - the CBOR text, it might be too late when you already have it in memory, - so you might want to check the size before you accept the string. - - Third, CBOR::XS recurses using the C stack when decoding objects and - arrays. The C stack is a limited resource: for instance, on my amd64 - machine with 8MB of stack size I can decode around 180k nested arrays - but only 14k nested CBOR objects (due to perl itself recursing deeply on - croak to free the temporary). If that is exceeded, the program crashes. - To be conservative, the default nesting limit is set to 512. If your - process has a smaller stack, you should adjust this setting accordingly - with the "max_depth" method. - - Something else could bomb you, too, that I forgot to think of. In that - case, you get to keep the pieces. I am always open for hints, though... - - Also keep in mind that CBOR::XS might leak contents of your Perl data - structures in its error messages, so when you serialise sensitive - information you might want to make sure that exceptions thrown by - CBOR::XS will not end up in front of untrusted eyes. + Disabling bigints will also partially or fully disable types that rely + on them, e.g. rational numbers that use bignums. CBOR IMPLEMENTATION NOTES This section contains some random implementation notes. They do not @@ -836,10 +1132,12 @@ LIMITATIONS ON PERLS WITHOUT 64-BIT INTEGER SUPPORT On perls that were built without 64 bit integer support (these are rare - nowadays, even on 32 bit architectures), support for any kind of 64 bit - integer in CBOR is very limited - most likely, these 64 bit values will - be truncated, corrupted, or otherwise not decoded correctly. This also - includes string, array and map sizes that are stored as 64 bit integers. + nowadays, even on 32 bit architectures, as all major Perl distributions + are built with 64 bit integer support), support for any kind of 64 bit + value in CBOR is very limited - most likely, these 64 bit values will be + truncated, corrupted, or otherwise not decoded correctly. This also + includes string, float, array and map sizes that are stored as 64 bit + integers. THREADS This module is *not* guaranteed to be thread safe and there are no plans