Quick DER, or if you like, "Quick and Easy DER", is a library for handling DER, which is a widely used binary representation of ASN.1 syntax in binary formats. The library describes ASN.1 syntax in a parser table that can be fed into library routines, resulting in pointer/length descriptors for the individual data fragments.
The basic approach of using this library is translating the ASN.1 syntax into a parser table. This is done when building your software, as a phase preceding the compilation of your Quick DER using program. The translation from ASN.1 to C code can be done manually and TODO will in the future be automated with an ASN.1 parser.
The resulting path walks are used in calls like der_unpack() to transform
DER into C-style structures, and der_pack() to make the opposite transformation.
The C-style structures are derived from the ASN.1 syntax, and permits access to
the information with no further need of understanding the depths of processing
DER.
Since you are handling binary data (rather than character strings), all data
is described in so-called dercursor structures; each containing a pointer
and a size of the data pointed at.
The output from mapping ASN.1 to a parser table is an include file for the
C programming language. This defines the various path walks that can be
used to unpack and pack DER data. The output and input of these routines
takes the form of an array of dercursor values.
To simplify use of the unpacked data, there are overlay structures for the
dercursor array. These overlay structures use the same labels that are
used in the ASN.1 syntax, so it is possible to walk around in the structures.
Some parts of the syntax indicate OPTIONAL elements. Such elements result
in the respective dercursor variables to be NULL values; specifically, the
function der_isnull() returns a true value for these elements.
There are routines der_iterate_first() and der_iterate_next() routines
to manually iterate over a DER structure's components. This can be used to
analyse structures that have not been unpacked (yet). The der_countelements()
routine can be used to predict the number of iterations.
There are also routines to manually walk through packaged DER structures,
namely der_enter() and der_skip() to enter into a nested structure and to
find the next element in a concatenation of such elements.
A much more advanced form of such walks through a DER structure exists in the
form of der_walk(), which is fed another kind of walk, a sequence of enter/skip
statements with tags that will be validated.
To validate the structures written in DER, both der_unpack() and der_walk()
can be used. Most other routines are coded for flexibility and should not be
assumed to validate DER in more detail than strictly required.
The der_unpack() routine runs through the entire structure, and validates
the tags it runs by, as well as the complete nesting structure it encounters.
It is a complete validation of the structures.
The der_walk() routine performs lazy validation, meaning that it will
carefully check tags and nesting inasfar as it is needed to get from its
starting point to its end point. Anything but the paths explored will be
accepted without question.
This is for ASN.1 experts; others can safely skip this subsection.
The Quick DER library is designed to process DER, although it will also accept
some of the more general BER format. If a length or value is written in more
than the minimal space, the library is still likely to accept it. Note that
the BIT STRING type is somewhat likely to run into overflow problems, so
there the full restrictiveness of DER is applied.
The entire library has been designed to operate without dynamic memory allocation. This means that there will never be a memory leak as a result of using Quick DER.
When DER information is unpacked, it is assumed to be loaded into a memory buffer
by the calling program, and the dercursor structures point to portions of that
buffer. The data is stored in dercursor arrays which the user program may
overlay with meaningful, ASN.1-labelled structures. In many applications, such
structures can be allocated on the stack.
Some portions of the data may be dynamically sized, notably the SEQUENCE OF
and SET OF structures, which indicate that the structural description following
it may be repeated in the binary data. Such data portions will be stored and
not yet unpacked by der_unpack(). Based on the stored DER data in a dercursor,
the calling application can choose to use iterators, der_walk() and so on to
avoid actually unpacking it; or it may allocate memory dynamically, and use that
to repeatedly call der_unpack() to find the individual entries.
In short, the Quick DER library never needs to perform memory allocation, and it provides the calling program with a lot of control to avoid it too. This is ideal for embedded applications, but is also beneficial for a secure programming style.
We have equiped a simple
ASN.1 Compiler
with a backend that generates both the overlay data structures and the
byte codes for der_pack() and der_unpack(). For most everyday ASN.1
this should suffice to produce a header file from an ASN.1 module.
We have started a collection of RFC's, stripped down to ASN.1 modules, that can be mapped to header files that can simply be include in a C program:
#include <quick-der/rfc5280.h>
#include <quick-der/rfc4120.h>
#include <quick-der/rfc4511.h>
These three examples were included already:
- X.509 certificate and CRL structures
- Kerberos packet formats and messaging
- LDAP as a protocol description
Most modules do things that confuse the compiler, so small changes have been
made, and clearly marked with asn1ate and an explanation in the ASN.1 files
in the rfc/ directory. The issues found were usually quite simple:
- we had to use uppercase initial characters in type names
- we removed complex constraints (as Quick DER does not implement them anyway)
- LDAP contained an infinitely circular reference for
Filter.not
There are some specifications that go far beyond the abilities of asn1ate,
by using such concepts as macros. This applied to the SNMP RFC's.
There are a few things that this library can use:
- More testing. The current
testdirectory is far too small; we can take a PKIX certificate apart and re-compose it, so we're definately doing something good, but this is nowhere near thorough testing. If you run into a problem case, then please share it so we can solve it and extend our test base. - Compiler output testing. The compiler output is currently compiled to see if there are C syntactical problems, but that is all. They have been visually compared to manually crafted code, too. More exhaustive tests, including a full application, would be in order.
- Python embedding. This would be useful for many protocol applications that need to modify a few things. It also greatly helps to get more protocols within the reach of Python. See our ideas about Python Quick DER for a plan that greatly simplifies ASN.1 processing with Python.
And of course, there are many useful things we may do with this library:
- Kerberos in PKIX. Certificates wrapping Kerberos Tickets for use with TLS-KDH
- Miniature LDAP services. These can help you centralise your data storage under own control; for instance, your PGP key ring or your vCard collection are good canidadates for sharing locally.
- Remote PKCS #11. The main issue in doing this well is proper encapsulation, but with Quick and Easy DER, and the emphasis of both on security and well-defined sizes, it appears to be a perfect match to wrap PKCS #11 arguments in DER structures.