2.6.1
=====
* [CVE-2013-1445] Fix PRNG not correctly reseeded in some situations.
In previous versions of PyCrypto, the Crypto.Random PRNG exhibits a
race condition that may cause forked processes to generate identical
sequences of 'random' numbers.
This is a fairly obscure bug that will (hopefully) not affect many
applications, but the failure scenario is pretty bad. Here is some
sample code that illustrates the problem:
from binascii import hexlify
import multiprocessing, pprint, time
import Crypto.Random
def task_main(arg):
a = Crypto.Random.get_random_bytes(8)
time.sleep(0.1)
b = Crypto.Random.get_random_bytes(8)
rdy, ack = arg
rdy.set()
ack.wait()
return "%s,%s" % (hexlify(a).decode(),
hexlify(b).decode())
n_procs = 4
manager = multiprocessing.Manager()
rdys = [manager.Event() for i in range(n_procs)]
acks = [manager.Event() for i in range(n_procs)]
Crypto.Random.get_random_bytes(1)
pool = multiprocessing.Pool(processes=n_procs,
initializer=Crypto.Random.atfork)
res_async = pool.map_async(task_main, zip(rdys, acks))
pool.close()
[rdy.wait() for rdy in rdys]
[ack.set() for ack in acks]
res = res_async.get()
pprint.pprint(sorted(res))
pool.join()
The output should be random, but it looked like this:
['c607803ae01aa8c0,2e4de6457a304b34',
'c607803ae01aa8c0,af80d08942b4c987',
'c607803ae01aa8c0,b0e4c0853de927c4',
'c607803ae01aa8c0,f0362585b3fceba4']
This release fixes the problem by resetting the rate-limiter when
Crypto.Random.atfork() is invoked. It also adds some tests and a
few related comments.
2.6
===
* [CVE-2012-2417] Fix LP#985164: insecure ElGamal key generation.
(thanks: Legrandin)
In the ElGamal schemes (for both encryption and signatures), g is
supposed to be the generator of the entire Z^*_p group. However, in
PyCrypto 2.5 and earlier, g is more simply the generator of a random
sub-group of Z^*_p.
The result is that the signature space (when the key is used for
signing) or the public key space (when the key is used for encryption)
may be greatly reduced from its expected size of log(p) bits, possibly
down to 1 bit (the worst case if the order of g is 2).
While it has not been confirmed, it has also been suggested that an
attacker might be able to use this fact to determine the private key.
Anyone using ElGamal keys should generate new keys as soon as practical.
Any additional information about this bug will be tracked at
https://bugs.launchpad.net/pycrypto/+bug/985164
* Huge documentation cleanup (thanks: Legrandin).
* Added more tests, including test vectors from NIST 800-38A
(thanks: Legrandin)
* Remove broken MODE_PGP, which never actually worked properly.
A new mode, MODE_OPENPGP, has been added for people wishing to write
OpenPGP implementations. Note that this does not implement the full
OpenPGP specification, only the "OpenPGP CFB mode" part of that
specification.
https://bugs.launchpad.net/pycrypto/+bug/996814
* Fix: getPrime with invalid input causes Python to abort with fatal error
https://bugs.launchpad.net/pycrypto/+bug/988431
* Fix: Segfaults within error-handling paths
(thanks: Paul Howarth & Dave Malcolm)
https://bugs.launchpad.net/pycrypto/+bug/934294
* Fix: Block ciphers allow empty string as IV
https://bugs.launchpad.net/pycrypto/+bug/997464
* Fix DevURandomRNG to work with Python3's new I/O stack.
(thanks: Sebastian Ramacher)
* Remove automagic dependencies on libgmp and libmpir, let the caller
disable them using args.
* Many other minor bug fixes and improvements (mostly thanks to Legrandin)
2.5
===
* Added PKCS#1 encryption schemes (v1.5 and OAEP). We now have
a decent, easy-to-use non-textbook RSA implementation. Yay!
* Added PKCS#1 signature schemes (v1.5 and PSS). v1.5 required some
extensive changes to Hash modules to contain the algorithm specific
ASN.1 OID. To that end, we now always have a (thin) Python module to
hide the one in pure C.
* Added 2 standard Key Derivation Functions (PBKDF1 and PBKDF2).
* Added export/import of RSA keys in OpenSSH and PKCS#8 formats.
* Added password-protected export/import of RSA keys (one old method
for PKCS#8 PEM only).
* Added ability to generate RSA key pairs with configurable public
exponent e.
* Added ability to construct an RSA key pair even if only the private
exponent d is known, and not p and q.
* Added SHA-2 C source code (fully from Lorenz Quack).
* Unit tests for all the above.
* Updates to documentation (both inline and in Doc/pycrypt.rst)
* All of the above changes were put together by Legrandin (Thanks!)
* Minor bug fixes (setup.py and tests).
2.4.1
=====
* Fix "error: Setup script exited with error: src/config.h: No such file or
directory" when installing via easy_install. (Sebastian Ramacher)
2.4
===
* Python 3 support! (Thorsten E. Behrens, Anders Sundman)
PyCrypto now supports every version of Python from 2.1 through 3.2.
* Timing-attack countermeasures in _fastmath: When built against
libgmp version 5 or later, we use mpz_powm_sec instead of mpz_powm.
This should prevent the timing attack described by Geremy Condra at
PyCon 2011:
http://blip.tv/pycon-us-videos-2009-2010-2011/pycon-2011-through-the-side-channel-timing-and-implementation-attacks-in-python-4897955
* New hash modules (for Python >= 2.5 only): SHA224, SHA384, and
SHA512 (Frédéric Bertolus)
* Configuration using GNU autoconf. This should help fix a bunch of
build issues.
* Support using MPIR as an alternative to GMP.
* Improve the test command in setup.py, by allowing tests to be
performed on a single sub-package or module only. (Legrandin)
You can now do something like this:
python setup.py test -m Hash.SHA256 --skip-slow-tests
* Fix double-decref of "counter" when Cipher object initialisation
fails (Ryan Kelly)
* Apply patches from Debian's python-crypto 2.3-3 package (Jan
Dittberner, Sebastian Ramacher):
- fix-RSA-generate-exception.patch
- epydoc-exclude-introspect.patch
- no-usr-local.patch
* Fix launchpad bug #702835: "Import key code is not compatible with
GMP library" (Legrandin)
* More tests, better documentation, various bugfixes.
2.3
===
* Fix NameError when attempting to use deprecated getRandomNumber()
function.
* _slowmath: Compute RSA u parameter when it's not given to
RSA.construct. This makes _slowmath behave the same as _fastmath in
this regard.
* Make RSA.generate raise a more user-friendly exception message when
the user tries to generate a bogus-length key.
2.2
===
* Deprecated Crypto.Util.number.getRandomNumber(), which had confusing
semantics. It's been replaced by getRandomNBitInteger and
getRandomInteger. (Thanks: Lorenz Quack)
* Better isPrime() and getPrime() implementations that do a real
Rabin-Miller probabilistic primality test (not the phony test we did
before with fixed bases). (Thanks: Lorenz Quack)
* getStrongPrime() implementation for generating RSA primes.
(Thanks: Lorenz Quack)
* Support for importing and exporting RSA keys in DER and PEM format.
(Thanks: Legrandin)
* Fix PyCrypto when floor division (python -Qnew) is enabled.
* When building using gcc, use -std=c99 for compilation. This should
fix building on FreeBSD and NetBSD.
2.1.0
=====
* Fix building PyCrypto on Win64 using MS Visual Studio 9.
(Thanks: Nevins Bartolomeo.)
2.1.0beta1
==========
* Modified RSA.generate() to ensure that e is coprime to p-1 and q-1.
Apparently, RSA.generate was capable of generating unusable keys.
2.1.0alpha2
===========
* Modified isPrime() to release the global interpreter lock while
performing computations. (patch from Lorenz Quack)
* Release the GIL while encrypting, decrypting, and hashing (but not
during initialization or finalization).
* API changes:
- Removed RandomPoolCompat and made Crypto.Util.randpool.RandomPool
a wrapper around Crypto.Random that emits a DeprecationWarning.
This is to discourage developers from attempting to provide
backwards compatibility for systems where there are NO strong
entropy sources available.
- Added Crypto.Random.get_random_bytes(). This should allow people
to use something like this if they want backwards-compatibility:
try:
from Crypto.Random import get_random_bytes
except ImportError:
try:
from os import urandom as get_random_bytes
except ImportError:
get_random_bytes = open("/dev/urandom", "rb").read
- Implemented __ne__() on pubkey, which fixes the following broken
behaviour:
>>> pk.publickey() == pk.publickey()
True
>>> pk.publickey() != pk.publickey()
True
(patch from Lorenz Quack)
- Block ciphers created with MODE_CTR can now operate on strings of
any size, rather than just multiples of the underlying cipher's
block size.
- Crypto.Util.Counter objects now raise OverflowError when they wrap
around to zero. You can override this new behaviour by passing
allow_wraparound=True to Counter.new()
2.1.0alpha1
===========
* This version supports Python versions 2.1 through 2.6.
* Clarified copyright status of much of the existing code by tracking
down Andrew M. Kuchling, Barry A. Warsaw, Jeethu Rao, Joris Bontje,
Mark Moraes, Paul Swartz, Robey Pointer, and Wim Lewis and getting
their permission to clarify the license/public-domain status of their
contributions. Many thanks to all involved!
* Replaced the test suite with a new, comprehensive package
(Crypto.SelfTest) that includes documentation about where its test
vectors came from, or how they were derived.
Use "python setup.py test" to run the tests after building.
* API changes:
- Added Crypto.version_info, which from now on will contain version
information in a format similar to Python's sys.version_info.
- Added a new random numbers API (Crypto.Random), and deprecated the
old one (Crypto.Util.randpool.RandomPool), which was misused more
often than not.
The new API is used by invoking Crypto.Random.new() and then just
reading from the file-like object that is returned.
CAVEAT: To maintain the security of the PRNG, you must call
Crypto.Random.atfork() in both the parent and the child processes
whenever you use os.fork(). Otherwise, the parent and child will
share copies of the same entropy pool, causing them to return the
same results! This is a limitation of Python, which does not
provide readily-accessible hooks to os.fork(). It's also a
limitation caused by the failure of operating systems to provide
sufficiently fast, trustworthy sources of cryptographically-strong
random numbers.
- Crypto.PublicKey now raises ValueError/TypeError/RuntimeError
instead of the various custom "error" exceptions
- Removed the IDEA and RC5 modules due to software patents. Debian
has been doing this for a while
- Added Crypto.Random.random, a strong version of the standard Python
'random' module.
- Added Crypto.Util.Counter, providing fast counter implementations
for use with CTR-mode ciphers.
* Bug fixes:
- Fixed padding bug in SHA256; this resulted in bad digests whenever
(the number of bytes hashed) mod 64 == 55.
- Fixed a 32-bit limitation on the length of messages the SHA256 module
could hash.
- AllOrNothing: Fixed padding bug in digest()
- Fixed a bad behaviour of the XOR cipher module: It would silently
truncate all keys to 32 bytes. Now it raises ValueError when the
key is too long.
- DSA: Added code to enforce FIPS 186-2 requirements on the size of
the prime p
- Fixed the winrandom module, which had been omitted from the build
process, causing security problems for programs that misuse RandomPool.
- Fixed infinite loop when attempting to generate RSA keys with an
odd number of bits in the modulus. (Not that you should do that.)
* Clarified the documentation for Crypto.Util.number.getRandomNumber.
Confusingly, this function does NOT return N random bits; It returns
a random N-bit number, i.e. a random number between 2**(N-1) and (2**N)-1.
Note that getRandomNumber is for internal use only and may be
renamed or removed in future releases.
* Replaced RIPEMD.c with a new implementation (RIPEMD160.c) to
alleviate copyright concerns.
* Replaced the DES/DES3 modules with ones based on libtomcrypt-1.16 to
alleviate copyright concerns.
* Replaced Blowfish.c with a new implementation to alleviate copyright
concerns.
* Added a string-XOR implementation written in C (Crypto.Util.strxor)
and used it to speed up Crypto.Hash.HMAC
* Converted documentation to reStructured Text.
* Added epydoc configuration Doc/epydoc-config
* setup.py now emits a warning when building without GMP.
* Added pct-speedtest.py to the source tree for doing performance
testing on the new code.
* Cleaned up the code in several places.
2.0.1
=====
* Fix SHA256 and RIPEMD on AMD64 platform.
* Deleted Demo/ directory.
* Add PublicKey to Crypto.__all__
2.0
===
* Added SHA256 module contributed by Jeethu Rao, with test data
from Taylor Boon.
* Fixed AES.c compilation problems with Borland C.
(Contributed by Jeethu Rao.)
* Fix ZeroDivisionErrors on Windows, caused by the system clock
not having enough resolution.
* Fix 2.1/2.2-incompatible use of (key not in dict),
pointed out by Ian Bicking.
* Fix FutureWarning in Crypto.Util.randpool, noted by James P Rutledge.
1.9alpha6
=========
* Util.number.getPrime() would inadvertently round off the bit
size; if you asked for a 129-bit prime or 135-bit prime, you
got a 128-bit prime.
* Added Util/test/prime_speed.py to measure the speed of prime
generation, and PublicKey/test/rsa_speed.py to measure
the speed of RSA operations.
* Merged the _rsa.c and _dsa.c files into a single accelerator
module, _fastmath.c.
* Speed improvements: Added fast isPrime() function to _fastmath,
cutting the time to generate a 1024-bit prime by a factor of 10.
Optimized the C version of RSA decryption to use a longer series
of operations that's roughly 3x faster than a single
exponentiation. (Contributed by Joris Bontje.)
* Added support to RSA key objects for blinding and unblinding
data. (Contributed by Joris Bontje.)
* Simplified RSA key generation: hard-wired the encryption
exponent to 65537 instead of generating a random prime;
generate prime factors in a loop until the product
is large enough.
* Renamed cansign(), canencrypt(), hasprivate(), to
can_sign, can_encrypt, has_private. If people shriek about
this change very loudly, I'll add aliases for the old method
names that log a warning and call the new method.
1.9alpha5
=========
* Many randpool changes. RandomPool now has a
randomize(N:int) method that can be called to get N
bytes of entropy for the pool (N defaults to 0,
which 'fills up' the pool's entropy) KeyboardRandom
overloads this method.
* Added src/winrand.c for Crypto.Util.winrandom and
now use winrandom for _randomize if possible.
(Calls Windows CryptoAPI CryptGenRandom)
* Several additional places for stirring the pool,
capturing inter-event entropy when reading/writing,
stirring before and after saves.
* RandomPool.add_event now returns the number of
estimated bits of added entropy, rather than the
pool entropy itself (since the pool entropy is
capped at the number of bits in the pool)
* Moved termios code from KeyboardRandomPool into a
KeyboardEntry class, provided a version for Windows
using msvcrt.
* Fix randpool.py crash on machines with poor timer resolution.
(Reported by Mark Moraes and others.)
* If the GNU GMP library is available, two C extensions will be
compiled to speed up RSA and DSA operations. (Contributed by
Paul Swartz.)
* DES3 with a 24-byte key was broken; now fixed.
(Patch by Philippe Frycia.)
1.9alpha4
=========
* Fix compilation problem on Windows.
* HMAC.py fixed to work with pre-2.2 Pythons
* setup.py now dies if built with Python 1.x
1.9alpha3
=========
* Fix a ref-counting bug that caused core dumps.
(Reported by Piers Lauder and an anonymous SF poster.)
1.9alpha2
=========
* (Backwards incompatible) The old Crypto.Hash.HMAC module is
gone, replaced by a copy of hmac.py from Python 2.2's standard
library. It will display a warning on interpreter versions
older than 2.2.
* (Backwards incompatible) Restored the Crypto.Protocol package,
and modernized and tidied up the two modules in it,
AllOrNothing.py and Chaffing.py, renaming various methods
and changing the interface.
* (Backwards incompatible) Changed the function names in
Crypto.Util.RFC1751.
* Restored the Crypto.PublicKey package at user request. I
think I'll leave it in the package and warn about it in the
documentation. I hope that eventually I can point to
someone else's better public-key code, and at that point I
may insert warnings and begin the process of deprecating
this code.
* Fix use of a Python 2.2 C function, replacing it with a
2.1-compatible equivalent. (Bug report and patch by Andrew
Eland.)
* Fix endianness bugs that caused test case failures on Sparc,
PPC, and doubtless other platforms.
* Fixed compilation problem on FreeBSD and MacOS X.
* Expanded the test suite (requires Sancho, from
http://www.mems-exchange.org/software/sancho/)
* Added lots of docstrings, so 'pydoc Crypto' now produces
helpful output. (Open question: maybe *all* of the documentation
should be moved into docstrings?)
* Make test.py automatically add the build/* directory to sys.path.
* Removed 'inline' declaration from C functions. Some compilers
don't support it, and Python's pyconfig.h no longer tells you whether
it's supported or not. After this change, some ciphers got slower,
but others got faster.
* The C-level API has been changed to reduce the amount of
memory-to-memory copying. This makes the code neater, but
had ambiguous performance effects; again, some ciphers got slower
and others became faster. Probably this is due to my compiler
optimizing slightly worse or better as a result.
* Moved C source implementations into src/ from block/, hash/,
and stream/. Having Hash/ and hash/ directories causes problems
on case-insensitive filesystems such as Mac OS.
* Cleaned up the C code for the extensions.
1.9alpha1
=========
* Added Crypto.Cipher.AES.
* Added the CTR mode and the variable-sized CFB mode from the
NIST standard on feedback modes.
* Removed Diamond, HAVAL, MD5, Sapphire, SHA, and Skipjack. MD5
and SHA are included with Python; the others are all of marginal
usefulness in the real world.
* Renamed the module-level constants ECB, CFB, &c., to MODE_ECB,
MODE_CFB, as part of making the block encryption modules
compliant with PEP 272. (I'm not sure about this change;
if enough users complain about it, I might back it out.)
* Made the hashing modules compliant with PEP 247 (not backward
compatible -- the major changes are that the constructor is now
MD2.new and not MD2.MD2, and the size of the digest is now
given as 'digest_size', not 'digestsize'.
* The Crypto.PublicKey package is no longer installed; the
interfaces are all wrong, and I have no idea what the right
interfaces should be.
1.1alpha2
=========
* Most importantly, the distribution has been broken into two
parts: exportable, and export-controlled. The exportable part
contains all the hashing algorithms, signature-only public key
algorithms, chaffing & winnowing, random number generation, various
utility modules, and the documentation.
The export-controlled part contains public-key encryption
algorithms such as RSA and ElGamal, and bulk encryption algorithms
like DES, IDEA, or Skipjack. Getting this code still requires that
you go through an access control CGI script, and denies you access if
you're outside the US or Canada.
* Added the RIPEMD hashing algorithm. (Contributed by
Hirendra Hindocha.)
* Implemented the recently declassified Skipjack block
encryption algorithm. My implementation runs at 864 K/sec on a
PII/266, which isn't particularly fast, but you're probably better off
using another algorithm anyway. :)
* A simple XOR cipher has been added, mostly for use by the
chaffing/winnowing code. (Contributed by Barry Warsaw.)
* Added Protocol.Chaffing and Hash.HMAC.py. (Contributed by
Barry Warsaw.)
Protocol.Chaffing implements chaffing and winnowing, recently
proposed by R. Rivest, which hides a message (the wheat) by adding
many noise messages to it (the chaff). The chaff can be discarded by
the receiver through a message authentication code. The neat thing
about this is that it allows secret communication without actually
having an encryption algorithm, and therefore this falls within the
exportable subset.
* Tidied up randpool.py, and removed its use of a block
cipher; this makes it work with only the export-controlled subset
available.
* Various renamings and reorganizations, mostly internal.
1.0.2
=====
* Changed files to work with Python 1.5; everything has been
re-arranged into a hierarchical package. (Not backward compatible.)
The package organization is:
Crypto.
Hash.
MD2, MD4, MD5, SHA, HAVAL
Cipher.
ARC2, ARC4, Blowfish, CAST, DES, DES3, Diamond,
IDEA, RC5, Sapphire
PublicKey.
DSA, ElGamal, qNEW, RSA
Util.
number, randpool, RFC1751
Since this is backward-incompatible anyway, I also changed
module names from all lower-case to mixed-case: diamond -> Diamond,
rc5 -> RC5, etc. That had been an annoying inconsistency for a while.
* Added CAST5 module contributed by <wiml@hhhh.org>.
* Added qNEW digital signature algorithm (from the digisign.py
I advertised a while back). (If anyone would like to suggest new
algorithms that should be implemented, please do; I think I've got
everything that's really useful at the moment, but...)
* Support for keyword arguments has been added. This allowed
removing the obnoxious key handling for Diamond and RC5, where the
first few bytes of the key indicated the number of rounds to use, and
various other parameters. Now you need only do something like:
from Crypto.Cipher import RC5
obj = RC5.new(key, RC5.ECB, rounds=8)
(Not backward compatible.)
* Various function names have been changed, and parameter
names altered. None of these were part of the public interface, so it
shouldn't really matter much.
* Various bugs fixed, the test suite has been expanded, and
the build process simplified.
* Updated the documentation accordingly.
1.0.1
=====
* Changed files to work with Python 1.4 .
* The DES and DES3 modules now automatically correct the
parity of their keys.
* Added R. Rivest's DES test (see http://theory.lcs.mit.edu/~rivest/destest.txt)
1.0.0
=====
* REDOC III succumbed to differential cryptanalysis, and has
been removed.
* The crypt and rotor modules have been dropped; they're still
available in the standard Python distribution.
* The Ultra-Fast crypt() module has been placed in a separate
distribution.
* Various bugs fixed.