This file contains the changes for the SSLeay library up to version 0.9.0b. For later changes, see the file "CHANGES". SSLeay CHANGES ______________ Changes between 0.8.x and 0.9.0b 10-Apr-1998 I said the next version would go out at easter, and so it shall. I expect a 0.9.1 will follow with portability fixes in the next few weeks. This is a quick, meet the deadline. Look to ssl-users for comments on what is new etc. eric (about to go bushwalking for the 4 day easter break :-) 16-Mar-98 - Patch for Cray T90 from Wayne Schroeder <schroede@SDSC.EDU> - Lots and lots of changes 29-Jan-98 - ASN1_BIT_STRING_set_bit()/ASN1_BIT_STRING_get_bit() from Goetz Babin-Ebell <babinebell@trustcenter.de>. - SSL_version() now returns SSL2_VERSION, SSL3_VERSION or TLS1_VERSION. 7-Jan-98 - Finally reworked the cipher string to ciphers again, so it works correctly - All the app_data stuff is now ex_data with funcion calls to access. The index is supplied by a function and 'methods' can be setup for the types that are called on XXX_new/XXX_free. This lets applications get notified on creation and destruction. Some of the RSA methods could be implemented this way and I may do so. - Oh yes, SSL under perl5 is working at the basic level. 15-Dec-97 - Warning - the gethostbyname cache is not fully thread safe, but it should work well enough. - Major internal reworking of the app_data stuff. More functions but if you were accessing ->app_data directly, things will stop working. - The perlv5 stuff is working. Currently on message digests, ciphers and the bignum library. 9-Dec-97 - Modified re-negotiation so that server initated re-neg will cause a SSL_read() to return -1 should retry. The danger otherwise was that the server and the client could end up both trying to read when using non-blocking sockets. 4-Dec-97 - Lots of small changes - Fix for binaray mode in Windows for the FILE BIO, thanks to Bob Denny <rdenny@dc3.com> 17-Nov-97 - Quite a few internal cleanups, (removal of errno, and using macros defined in e_os.h). - A bug in ca.c, pointed out by yasuyuki-ito@d-cruise.co.jp, where the automactic naming out output files was being stuffed up. 29-Oct-97 - The Cast5 cipher has been added. MD5 and SHA-1 are now in assember for x86. 21-Oct-97 - Fixed a bug in the BIO_gethostbyname() cache. 15-Oct-97 - cbc mode for blowfish/des/3des is now in assember. Blowfish asm has also been improved. At this point in time, on the pentium, md5 is %80 faster, the unoptimesed sha-1 is %79 faster, des-cbc is %28 faster, des-ede3-cbc is %9 faster and blowfish-cbc is %62 faster. 12-Oct-97 - MEM_BUF_grow() has been fixed so that it always sets the buf->length to the value we are 'growing' to. Think of MEM_BUF_grow() as the way to set the length value correctly. 10-Oct-97 - I now hash for certificate lookup on the raw DER encoded RDN (md5). This breaks things again :-(. This is efficent since I cache the DER encoding of the RDN. - The text DN now puts in the numeric OID instead of UNKNOWN. - req can now process arbitary OIDs in the config file. - I've been implementing md5 in x86 asm, much faster :-). - Started sha1 in x86 asm, needs more work. - Quite a few speedups in the BN stuff. RSA public operation has been made faster by caching the BN_MONT_CTX structure. The calulating of the Ai where A*Ai === 1 mod m was rather expensive. Basically a 40-50% speedup on public operations. The RSA speedup is now 15% on pentiums and %20 on pentium pro. 30-Sep-97 - After doing some profiling, I added x86 adm for bn_add_words(), which just adds 2 arrays of longs together. A %10 speedup for 512 and 1024 bit RSA on the pentium pro. 29-Sep-97 - Converted the x86 bignum assembler to us the perl scripts for generation. 23-Sep-97 - If SSL_set_session() is passed a NULL session, it now clears the current session-id. 22-Sep-97 - Added a '-ss_cert file' to apps/ca.c. This will sign selfsigned certificates. - Bug in crypto/evp/encode.c where by decoding of 65 base64 encoded lines, one line at a time (via a memory BIO) would report EOF after the first line was decoded. - Fix in X509_find_by_issuer_and_serial() from Dr Stephen Henson <shenson@bigfoot.com> 19-Sep-97 - NO_FP_API and NO_STDIO added. - Put in sh config command. It auto runs Configure with the correct parameters. 18-Sep-97 - Fix x509.c so if a DSA cert has different parameters to its parent, they are left in place. Not tested yet. 16-Sep-97 - ssl_create_cipher_list() had some bugs, fixes from Patrick Eisenacher <eisenach@stud.uni-frankfurt.de> - Fixed a bug in the Base64 BIO, where it would return 1 instead of -1 when end of input was encountered but should retry. Basically a Base64/Memory BIO interaction problem. - Added a HMAC set of functions in preporarion for TLS work. 15-Sep-97 - Top level makefile tweak - Cameron Simpson <cs@zip.com.au> - Prime generation spead up %25 (512 bit prime, pentium pro linux) by using montgomery multiplication in the prime number test. 11-Sep-97 - Ugly bug in ssl3_write_bytes(). Basically if application land does a SSL_write(ssl,buf,len) where len > 16k, the SSLv3 write code did not check the size and tried to copy the entire buffer. This would tend to cause memory overwrites since SSLv3 has a maximum packet size of 16k. If your program uses buffers <= 16k, you would probably never see this problem. - Fixed a new errors that were cause by malloc() not returning 0 initialised memory.. - SSL_OP_NETSCAPE_CA_DN_BUG was being switched on when using SSL_CTX_set_options(ssl_ctx,SSL_OP_ALL); which was a bad thing since this flags stops SSLeay being able to handle client cert requests correctly. 08-Sep-97 - SSL_SESS_CACHE_NO_INTERNAL_LOOKUP option added. When switched on, the SSL server routines will not use a SSL_SESSION that is held in it's cache. This in intended to be used with the session-id callbacks so that while the session-ids are still stored in the cache, the decision to use them and how to look them up can be done by the callbacks. The are the 'new', 'get' and 'remove' callbacks. This can be used to determine the session-id to use depending on information like which port/host the connection is coming from. Since the are also SSL_SESSION_set_app_data() and SSL_SESSION_get_app_data() functions, the application can hold information against the session-id as well. 03-Sep-97 - Added lookup of CRLs to the by_dir method, X509_load_crl_file() also added. Basically it means you can lookup CRLs via the same system used to lookup certificates. - Changed things so that the X509_NAME structure can contain ASN.1 BIT_STRINGS which is required for the unique identifier OID. - Fixed some problems with the auto flushing of the session-id cache. It was not occuring on the server side. 02-Sep-97 - Added SSL_CTX_sess_cache_size(SSL_CTX *ctx,unsigned long size) which is the maximum number of entries allowed in the session-id cache. This is enforced with a simple FIFO list. The default size is 20*1024 entries which is rather large :-). The Timeout code is still always operating. 01-Sep-97 - Added an argument to all the 'generate private key/prime` callbacks. It is the last parameter so this should not break existing code but it is needed for C++. - Added the BIO_FLAGS_BASE64_NO_NL flag for the BIO_f_base64() BIO. This lets the BIO read and write base64 encoded data without inserting or looking for '\n' characters. The '-A' flag turns this on when using apps/enc.c. - RSA_NO_PADDING added to help BSAFE functionality. This is a very dangerous thing to use, since RSA private key operations without random padding bytes (as PKCS#1 adds) can be attacked such that the private key can be revealed. - ASN.1 bug and rc2-40-cbc and rc4-40 added by Dr Stephen Henson <shenson@bigfoot.com> 31-Aug-97 (stuff added while I was away) - Linux pthreads by Tim Hudson (tjh@cryptsoft.com). - RSA_flags() added allowing bypass of pub/priv match check in ssl/ssl_rsa.c - Tim Hudson. - A few minor bugs. SSLeay 0.8.1 released. 19-Jul-97 - Server side initated dynamic renegotiation is broken. I will fix it when I get back from holidays. 15-Jul-97 - Quite a few small changes. - INVALID_SOCKET usage cleanups from Alex Kiernan <alex@hisoft.co.uk> 09-Jul-97 - Added 2 new values to the SSL info callback. SSL_CB_START which is passed when the SSL protocol is started and SSL_CB_DONE when it has finished sucsessfully. 08-Jul-97 - Fixed a few bugs problems in apps/req.c and crypto/asn1/x_pkey.c that related to DSA public/private keys. - Added all the relevent PEM and normal IO functions to support reading and writing RSAPublic keys. - Changed makefiles to use ${AR} instead of 'ar r' 07-Jul-97 - Error in ERR_remove_state() that would leave a dangling reference to a free()ed location - thanks to Alex Kiernan <alex@hisoft.co.uk> - s_client now prints the X509_NAMEs passed from the server when requesting a client cert. - Added a ssl->type, which is one of SSL_ST_CONNECT or SSL_ST_ACCEPT. I had to add it so I could tell if I was a connect or an accept after the handshake had finished. - SSL_get_client_CA_list(SSL *s) now returns the CA names passed by the server if called by a client side SSL. 05-Jul-97 - Bug in X509_NAME_get_text_by_OBJ(), looking starting at index 0, not -1 :-( Fix from Tim Hudson (tjh@cryptsoft.com). 04-Jul-97 - Fixed some things in X509_NAME_add_entry(), thanks to Matthew Donald <matthew@world.net>. - I had a look at the cipher section and though that it was a bit confused, so I've changed it. - I was not setting up the RC4-64-MD5 cipher correctly. It is a MS special that appears in exported MS Money. - Error in all my DH ciphers. Section 7.6.7.3 of the SSLv3 spec. I was missing the two byte length header for the ClientDiffieHellmanPublic value. This is a packet sent from the client to the server. The SSL_OP_SSLEAY_080_CLIENT_DH_BUG option will enable SSLeay server side SSLv3 accept either the correct or my 080 packet format. - Fixed a few typos in crypto/pem.org. 02-Jul-97 - Alias mapping for EVP_get_(digest|cipher)byname is now performed before a lookup for actual cipher. This means that an alias can be used to 're-direct' a cipher or a digest. - ASN1_read_bio() had a bug that only showed up when using a memory BIO. When EOF is reached in the memory BIO, it is reported as a -1 with BIO_should_retry() set to true. 01-Jul-97 - Fixed an error in X509_verify_cert() caused by my miss-understanding how 'do { contine } while(0);' works. Thanks to Emil Sit <sit@mit.edu> for educating me :-) 30-Jun-97 - Base64 decoding error. If the last data line did not end with a '=', sometimes extra data would be returned. - Another 'cut and paste' bug in x509.c related to setting up the STDout BIO. 27-Jun-97 - apps/ciphers.c was not printing due to an editing error. - Alex Kiernan <alex@hisoft.co.uk> send in a nice fix for a library build error in util/mk1mf.pl 26-Jun-97 - Still did not have the auto 'experimental' code removal script correct. - A few header tweaks for Watcom 11.0 under Win32 from Rolf Lindemann <Lindemann@maz-hh.de> - 0 length OCTET_STRING bug in asn1_parse - A minor fix with an non-existent function in the MS .def files. - A few changes to the PKCS7 stuff. 25-Jun-97 SSLeay 0.8.0 finally it gets released. 24-Jun-97 Added a SSL_OP_EPHEMERAL_RSA option which causes all SSLv3 RSA keys to use a temporary RSA key. This is experimental and needs some more work. Fixed a few Win16 build problems. 23-Jun-97 SSLv3 bug. I was not doing the 'lookup' of the CERT structure correctly. I was taking the SSL->ctx->default_cert when I should have been using SSL->cert. The bug was in ssl/s3_srvr.c 20-Jun-97 X509_ATTRIBUTES were being encoded wrongly by apps/reg.c and the rest of the library. Even though I had the code required to do it correctly, apps/req.c was doing the wrong thing. I have fixed and tested everything. Missing a few #ifdef FIONBIO sections in crypto/bio/bss_acpt.c. 19-Jun-97 Fixed a bug in the SSLv2 server side first packet handling. When using the non-blocking test BIO, the ssl->s2->first_packet flag was being reset when a would-block failure occurred when reading the first 5 bytes of the first packet. This caused the checking logic to run at the wrong time and cause an error. Fixed a problem with specifying cipher. If RC4-MD5 were used, only the SSLv3 version would be picked up. Now this will pick up both SSLv2 and SSLv3 versions. This required changing the SSL_CIPHER->mask values so that they only mask the ciphers, digests, authentication, export type and key-exchange algorithms. I found that when a SSLv23 session is established, a reused session, of type SSLv3 was attempting to write the SSLv2 ciphers, which were invalid. The SSL_METHOD->put_cipher_by_char method has been modified so it will only write out cipher which that method knows about. Changes between 0.8.0 and 0.8.1 *) Mostly bug fixes. There is an Ephemeral DH cipher problem which is fixed. SSLeay 0.8.0 This version of SSLeay has quite a lot of things different from the previous version. Basically check all callback parameters, I will be producing documentation about how to use things in th future. Currently I'm just getting 080 out the door. Please not that there are several ways to do everything, and most of the applications in the apps directory are hybrids, some using old methods and some using new methods. Have a look in demos/bio for some very simple programs and apps/s_client.c and apps/s_server.c for some more advanced versions. Notes are definitly needed but they are a week or so away. Anyway, some quick nots from Tim Hudson (tjh@cryptsoft.com) --- Quick porting notes for moving from SSLeay-0.6.x to SSLeay-0.8.x to get those people that want to move to using the new code base off to a quick start. Note that Eric has tidied up a lot of the areas of the API that were less than desirable and renamed quite a few things (as he had to break the API in lots of places anyrate). There are a whole pile of additional functions for making dealing with (and creating) certificates a lot cleaner. 01-Jul-97 Tim Hudson tjh@cryptsoft.com ---8<--- To maintain code that uses both SSLeay-0.6.x and SSLeay-0.8.x you could use something like the following (assuming you #include "crypto.h" which is something that you really should be doing). #if SSLEAY_VERSION_NUMBER >= 0x0800 #define SSLEAY8 #endif buffer.h -> splits into buffer.h and bio.h so you need to include bio.h too if you are working with BIO internal stuff (as distinct from simply using the interface in an opaque manner) #include "bio.h" - required along with "buffer.h" if you write your own BIO routines as the buffer and bio stuff that was intermixed has been separated out envelope.h -> evp.h (which should have been done ages ago) Initialisation ... don't forget these or you end up with code that is missing the bits required to do useful things (like ciphers): SSLeay_add_ssl_algorithms() (probably also want SSL_load_error_strings() too but you should have already had that call in place) SSL_CTX_new() - requires an extra method parameter SSL_CTX_new(SSLv23_method()) SSL_CTX_new(SSLv2_method()) SSL_CTX_new(SSLv3_method()) OR to only have the server or the client code SSL_CTX_new(SSLv23_server_method()) SSL_CTX_new(SSLv2_server_method()) SSL_CTX_new(SSLv3_server_method()) or SSL_CTX_new(SSLv23_client_method()) SSL_CTX_new(SSLv2_client_method()) SSL_CTX_new(SSLv3_client_method()) SSL_set_default_verify_paths() ... renamed to the more appropriate SSL_CTX_set_default_verify_paths() If you want to use client certificates then you have to add in a bit of extra stuff in that a SSLv3 server sends a list of those CAs that it will accept certificates from ... so you have to provide a list to SSLeay otherwise certain browsers will not send client certs. SSL_CTX_set_client_CA_list(ctx,SSL_load_client_CA_file(s_cert_file)); X509_NAME_oneline(X) -> X509_NAME_oneline(X,NULL,0) or provide a buffer and size to copy the result into X509_add_cert -> X509_STORE_add_cert (and you might want to read the notes on X509_NAME structure changes too) VERIFICATION CODE ================= The codes have all be renamed from VERIFY_ERR_* to X509_V_ERR_* to more accurately reflect things. The verification callback args are now packaged differently so that extra fields for verification can be added easily in future without having to break things by adding extra parameters each release :-) X509_cert_verify_error_string -> X509_verify_cert_error_string BIO INTERNALS ============= Eric has fixed things so that extra flags can be introduced in the BIO layer in future without having to play with all the BIO modules by adding in some macros. The ugly stuff using b->flags ~= (BIO_FLAGS_RW|BIO_FLAGS_SHOULD_RETRY) becomes BIO_clear_retry_flags(b) b->flags |= (BIO_FLAGS_READ|BIO_FLAGS_SHOULD_RETRY) becomes BIO_set_retry_read(b) Also ... BIO_get_retry_flags(b), BIO_set_flags(b) OTHER THINGS ============ X509_NAME has been altered so that it isn't just a STACK ... the STACK is now in the "entries" field ... and there are a pile of nice functions for getting at the details in a much cleaner manner. SSL_CTX has been altered ... "cert" is no longer a direct member of this structure ... things are now down under "cert_store" (see x509_vfy.h) and things are no longer in a CERTIFICATE_CTX but instead in a X509_STORE. If your code "knows" about this level of detail then it will need some surgery. If you depending on the incorrect spelling of a number of the error codes then you will have to change your code as these have been fixed. ENV_CIPHER "type" got renamed to "nid" and as that is what it actually has been all along so this makes things clearer. ify_cert_error_string(ctx->error)); SSL_R_NO_CIPHER_WE_TRUST -> SSL_R_NO_CIPHER_LIST and SSL_R_REUSE_CIPHER_LIST_NOT_ZERO Changes between 0.7.x and 0.8.0 *) There have been lots of changes, mostly the addition of SSLv3. There have been many additions from people and amongst others, C2Net has assisted greatly. Changes between 0.7.x and 0.7.x *) Internal development version only SSLeay 0.6.6 13-Jan-1997 The main additions are - assember for x86 DES improvments. From 191,000 per second on a pentium 100, I now get 281,000. The inner loop and the IP/FP modifications are from Svend Olaf Mikkelsen <svolaf@inet.uni-c.dk>. Many thanks for his contribution. - The 'DES macros' introduced in 0.6.5 now have 3 types. DES_PTR1, DES_PTR2 and 'normal'. As per before, des_opts reports which is best and there is a summery of mine in crypto/des/options.txt - A few bug fixes. - Added blowfish. It is not used by SSL but all the other stuff that deals with ciphers can use it in either ecb, cbc, cfb64 or ofb64 modes. There are 3 options for optimising Blowfish. BF_PTR, BF_PTR2 and 'normal'. BF_PTR2 is pentium/x86 specific. The correct option is setup in the 'Configure' script. - There is now a 'get client certificate' callback which can be 'non-blocking'. If more details are required, let me know. It will documented more in SSLv3 when I finish it. - Bug fixes from 0.6.5 including the infamous 'ca' bug. The 'make test' now tests the ca program. - Lots of little things modified and tweaked. SSLeay 0.6.5 After quite some time (3 months), the new release. I have been very busy for the last few months and so this is mostly bug fixes and improvments. The main additions are - assember for x86 DES. For all those gcc based systems, this is a big improvement. From 117,000 DES operation a second on a pentium 100, I now get 191,000. I have also reworked the C version so it now gives 148,000 DESs per second. - As mentioned above, the inner DES macros now have some more variant that sometimes help, sometimes hinder performance. There are now 3 options DES_PTR (ptr vs array lookup), DES_UNROLL (full vs partial loop unrolling) and DES_RISC (a more register intensive version of the inner macro). The crypto/des/des_opts.c program, when compiled and run, will give an indication of the correct options to use. - The BIO stuff has been improved. Read doc/bio.doc. There are now modules for encryption and base64 encoding and a BIO_printf() function. - The CA program will accept simple one line X509v3 extensions in the ssleay.cnf file. Have a look at the example. Currently this just puts the text into the certificate as an OCTET_STRING so currently the more advanced X509v3 data types are not handled but this is enough for the netscape extensions. - There is the start of a nicer higher level interface to the X509 strucutre. - Quite a lot of bug fixes. - CRYPTO_malloc_init() (or CRYPTO_set_mem_functions()) can be used to define the malloc(), free() and realloc() routines to use (look in crypto/crypto.h). This is mostly needed for Windows NT/95 when using DLLs and mixing CRT libraries. In general, read the 'VERSION' file for changes and be aware that some of the new stuff may not have been tested quite enough yet, so don't just plonk in SSLeay 0.6.5 when 0.6.4 used to work and expect nothing to break. SSLeay 0.6.4 30/08/96 eay I've just finished some test builds on Windows NT, Windows 3.1, Solaris 2.3, Solaris 2.5, Linux, IRIX, HPUX 10 and everthing seems to work :-). The main changes in this release - Thread safe. have a read of doc/threads.doc and play in the mt directory. For anyone using 0.6.3 with threads, I found 2 major errors so consider moving to 0.6.4. I have a test program that builds under NT and solaris. - The get session-id callback has changed. Have a read of doc/callback.doc. - The X509_cert_verify callback (the SSL_verify callback) now has another argument. Have a read of doc/callback.doc - 'ca -preserve', sign without re-ordering the DN. Not tested much. - VMS support. - Compile time memory leak detection can now be built into SSLeay. Read doc/memory.doc - CONF routines now understand '\', '\n', '\r' etc. What this means is that the SPKAC object mentioned in doc/ns-ca.doc can be on multiple lines. - 'ssleay ciphers' added, lists the default cipher list for SSLeay. - RC2 key setup is now compatable with Netscape. - Modifed server side of SSL implementation, big performance difference when using session-id reuse. 0.6.3 Bug fixes and the addition of some nice stuff to the 'ca' program. Have a read of doc/ns-ca.doc for how hit has been modified so it can be driven from a CGI script. The CGI script is not provided, but that is just being left as an excersize for the reader :-). 0.6.2 This is most bug fixes and functionality improvements. Additions are - More thread debugging patches, the thread stuff is still being tested, but for those keep to play with stuff, have a look in crypto/cryptlib.c. The application needs to define 1 (or optionaly a second) callback that is used to implement locking. Compiling with LOCK_DEBUG spits out lots of locking crud :-). This is what I'm currently working on. - SSL_CTX_set_default_passwd_cb() can be used to define the callback function used in the SSL*_file() functions used to load keys. I was always of the opinion that people should call PEM_read_RSAPrivateKey() and pass the callback they want to use, but it appears they just want to use the SSL_*_file() function() :-(. - 'enc' now has a -kfile so a key can be read from a file. This is mostly used so that the passwd does not appear when using 'ps', which appears imposible to stop under solaris. - X509v3 certificates now work correctly. I even have more examples in my tests :-). There is now a X509_EXTENSION type that is used in X509v3 certificates and CRLv2. - Fixed that signature type error :-( - Fixed quite a few potential memory leaks and problems when reusing X509, CRL and REQ structures. - EVP_set_pw_prompt() now sets the library wide default password prompt. - The 'pkcs7' command will now, given the -print_certs flag, output in pem format, all certificates and CRL contained within. This is more of a pre-emtive thing for the new verisign distribution method. I should also note, that this also gives and example in code, of how to do this :-), or for that matter, what is involved in going the other way (list of certs and crl -> pkcs7). - Added RSA's DESX to the DES library. It is also available via the EVP_desx_cbc() method and via 'enc desx'. SSLeay 0.6.1 The main functional changes since 0.6.0 are as follows - Bad news, the Microsoft 060 DLL's are not compatable, but the good news is that from now on, I'll keep the .def numbers the same so they will be. - RSA private key operations are about 2 times faster that 0.6.0 - The SSL_CTX now has more fields so default values can be put against it. When an SSL structure is created, these default values are used but can be overwritten. There are defaults for cipher, certificate, private key, verify mode and callback. This means SSL session creation can now be ssl=SSL_new() SSL_set_fd(ssl,sock); SSL_accept(ssl) .... All the other uglyness with having to keep a global copy of the private key and certificate/verify mode in the server is now gone. - ssl/ssltest.c - one process talking SSL to its self for testing. - Storage of Session-id's can be controled via a session_cache_mode flag. There is also now an automatic default flushing of old session-id's. - The X509_cert_verify() function now has another parameter, this should not effect most people but it now means that the reason for the failure to verify is now available via SSL_get_verify_result(ssl). You don't have to use a global variable. - SSL_get_app_data() and SSL_set_app_data() can be used to keep some application data against the SSL structure. It is upto the application to free the data. I don't use it, but it is available. - SSL_CTX_set_cert_verify_callback() can be used to specify a verify callback function that completly replaces my certificate verification code. Xcert should be able to use this :-). The callback is of the form int app_verify_callback(arg,ssl,cert). This needs to be documented more. - I have started playing with shared library builds, have a look in the shlib directory. It is very simple. If you need a numbered list of functions, have a look at misc/crypto.num and misc/ssl.num. - There is some stuff to do locking to make the library thread safe. I have only started this stuff and have not finished. If anyone is keen to do so, please send me the patches when finished. So I have finally made most of the additions to the SSL interface that I thought were needed. There will probably be a pause before I make any non-bug/documentation related changes to SSLeay since I'm feeling like a bit of a break. eric - 12 Jul 1996 I saw recently a comment by some-one that we now seem to be entering the age of perpetual Beta software. Pioneered by packages like linux but refined to an art form by netscape. I too wish to join this trend with the anouncement of SSLeay 0.6.0 :-). There are quite a large number of sections that are 'works in progress' in this package. I will also list the major changes and what files you should read. BIO - this is the new IO structure being used everywhere in SSLeay. I started out developing this because of microsoft, I wanted a mechanism to callback to the application for all IO, so Windows 3.1 DLL perversion could be hidden from me and the 15 different ways to write to a file under NT would also not be dictated by me at library build time. What the 'package' is is an API for a data structure containing functions. IO interfaces can be written to conform to the specification. This in not intended to hide the underlying data type from the application, but to hide it from SSLeay :-). I have only really finished testing the FILE * and socket/fd modules. There are also 'filter' BIO's. Currently I have only implemented message digests, and it is in use in the dgst application. This functionality will allow base64/encrypto/buffering modules to be 'push' into a BIO without it affecting the semantics. I'm also working on an SSL BIO which will hide the SSL_accept()/SLL_connet() from an event loop which uses the interface. It is also possible to 'attach' callbacks to a BIO so they get called before and after each operation, alowing extensive debug output to be generated (try running dgst with -d). Unfortunaly in the conversion from 0.5.x to 0.6.0, quite a few functions that used to take FILE *, now take BIO *. The wrappers are easy to write function_fp(fp,x) FILE *fp; { BIO *b; int ret; if ((b=BIO_new(BIO_s_file())) == NULL) error..... BIO_set_fp(b,fp,BIO_NOCLOSE); ret=function_bio(b,x); BIO_free(b); return(ret); } Remember, there are no functions that take FILE * in SSLeay when compiled for Windows 3.1 DLL's. -- I have added a general EVP_PKEY type that can hold a public/private key. This is now what is used by the EVP_ functions and is passed around internally. I still have not done the PKCS#8 stuff, but X509_PKEY is defined and waiting :-) -- For a full function name listings, have a look at ms/crypt32.def and ms/ssl32.def. These are auto-generated but are complete. Things like ASN1_INTEGER_get() have been added and are in here if you look. I have renamed a few things, again, have a look through the function list and you will probably find what you are after. I intend to at least put a one line descrition for each one..... -- Microsoft - thats what this release is about, read the MICROSOFT file. -- Multi-threading support. I have started hunting through the code and flaging where things need to be done. In a state of work but high on the list. -- For random numbers, edit e_os.h and set DEVRANDOM (it's near the top) be be you random data device, otherwise 'RFILE' in e_os.h will be used, in your home directory. It will be updated periodically. The environment variable RANDFILE will override this choice and read/write to that file instead. DEVRANDOM is used in conjunction to the RFILE/RANDFILE. If you wish to 'seed' the random number generator, pick on one of these files. -- The list of things to read and do dgst -d s_client -state (this uses a callback placed in the SSL state loop and will be used else-where to help debug/monitor what is happening.) doc/why.doc doc/bio.doc <- hmmm, needs lots of work. doc/bss_file.doc <- one that is working :-) doc/session.doc <- it has changed doc/speed.doc also play with ssleay version -a. I have now added a SSLeay() function that returns a version number, eg 0600 for this release which is primarily to be used to check DLL version against the application. util/* Quite a few will not interest people, but some may, like mk1mf.pl, mkdef.pl, util/do_ms.sh try cc -Iinclude -Icrypto -c crypto/crypto.c cc -Iinclude -Issl -c ssl/ssl.c You have just built the SSLeay libraries as 2 object files :-) Have a general rummage around in the bin stall directory and look at what is in there, like CA.sh and c_rehash There are lots more things but it is 12:30am on a Friday night and I'm heading home :-). eric 22-Jun-1996 This version has quite a few major bug fixes and improvements. It DOES NOT do SSLv3 yet. The main things changed - A Few days ago I added the s_mult application to ssleay which is a demo of an SSL server running in an event loop type thing. It supports non-blocking IO, I have finally gotten it right, SSL_accept() can operate in non-blocking IO mode, look at the code to see how :-). Have a read of doc/s_mult as well. This program leaks memory and file descriptors everywhere but I have not cleaned it up yet. This is a demo of how to do non-blocking IO. - The SSL session management has been 'worked over' and there is now quite an expansive set of functions to manipulate them. Have a read of doc/session.doc for some-things I quickly whipped up about how it now works. This assume you know the SSLv2 protocol :-) - I can now read/write the netscape certificate format, use the -inform/-outform 'net' options to the x509 command. I have not put support for this type in the other demo programs, but it would be easy to add. - asn1parse and 'enc' have been modified so that when reading base64 encoded files (pem format), they do not require '-----BEGIN' header lines. The 'enc' program had a buffering bug fixed, it can be used as a general base64 -> binary -> base64 filter by doing 'enc -a -e' and 'enc -a -d' respecivly. Leaving out the '-a' flag in this case makes the 'enc' command into a form of 'cat'. - The 'x509' and 'req' programs have been fixed and modified a little so that they generate self-signed certificates correctly. The test script actually generates a 'CA' certificate and then 'signs' a 'user' certificate. Have a look at this shell script (test/sstest) to see how things work, it tests most possible combinations of what can be done. - The 'SSL_set_pref_cipher()' function has been 'fixed' and the prefered name of SSL_set_cipher_list() is now the correct API (stops confusion :-). If this function is used in the client, only the specified ciphers can be used, with preference given to the order the ciphers were listed. For the server, if this is used, only the specified ciphers will be used to accept connections. If this 'option' is not used, a default set of ciphers will be used. The SSL_CTX_set_cipher_list(SSL_CTX *ctx) sets this list for all ciphers started against the SSL_CTX. So the order is SSL cipher_list, if not present, SSL_CTX cipher list, if not present, then the library default. What this means is that normally ciphers like NULL-MD5 will never be used. The only way this cipher can be used for both ends to specify to use it. To enable or disable ciphers in the library at build time, modify the first field for the cipher in the ssl_ciphers array in ssl/ssl_lib.c. This file also contains the 'pref_cipher' list which is the default cipher preference order. - I'm not currently sure if the 'rsa -inform net' and the 'rsa -outform net' options work. They should, and they enable loading and writing the netscape rsa private key format. I will be re-working this section of SSLeay for the next version. What is currently in place is a quick and dirty hack. - I've re-written parts of the bignum library. This gives speedups for all platforms. I now provide assembler for use under Windows NT. I have not tested the Windows 3.1 assembler but it is quite simple code. This gives RSAprivate_key operation encryption times of 0.047s (512bit key) and 0.230s (1024bit key) on a pentium 100 which I consider reasonable. Basically the times available under linux/solaris x86 can be achieve under Windows NT. I still don't know how these times compare to RSA's BSAFE library but I have been emailing with people and with their help, I should be able to get my library's quite a bit faster still (more algorithm changes). The object file crypto/bn/asm/x86-32.obj should be used when linking under NT. - 'make makefile.one' in the top directory will generate a single makefile called 'makefile.one' This makefile contains no perl references and will build the SSLeay library into the 'tmp' and 'out' directories. util/mk1mf.pl >makefile.one is how this makefile is generated. The mk1mf.pl command take several option to generate the makefile for use with cc, gcc, Visual C++ and Borland C++. This is still under development. I have only build .lib's for NT and MSDOS I will be working on this more. I still need to play with the correct compiler setups for these compilers and add some more stuff but basically if you just want to compile the library on a 'non-unix' platform, this is a very very good file to start with :-). Have a look in the 'microsoft' directory for my current makefiles. I have not yet modified things to link with sockets under Windows NT. You guys should be able to do this since this is actually outside of the SSLeay scope :-). I will be doing it for myself soon. util/mk1mf.pl takes quite a few options including no-rc, rsaref and no-sock to build without RC2/RC4, to require RSAref for linking, and to build with no socket code. - Oh yes, the cipher that was reported to be compatible with RSA's RC2 cipher that was posted to sci.crypt has been added to the library and SSL. I take the view that if RC2 is going to be included in a standard, I'll include the cipher to make my package complete. There are NO_RC2, NO_RC4 and NO_IDEA macros to remove these ciphers at compile time. I have not tested this recently but it should all work and if you are in the USA and don't want RSA threatening to sue you, you could probably remove the RC4/RC2 code inside these sections. I may in the future include a perl script that does this code removal automatically for those in the USA :-). - I have removed all references to sed in the makefiles. So basically, the development environment requires perl and sh. The build environment does not (use the makefile.one makefile). The Configure script still requires perl, this will probably stay that way since I have perl for Windows NT :-). eric (03-May-1996) PS Have a look in the VERSION file for more details on the changes and bug fixes. I have fixed a few bugs, added alpha and x86 assembler and generally cleaned things up. This version will be quite stable, mostly because I'm on holidays until 10-March-1996. For any problems in the interum, send email to Tim Hudson <tjh@mincom.oz.au>. SSLeay 0.5.0 12-12-95 This is going out before it should really be released. I leave for 11 weeks holidays on the 22-12-95 and so I either sit on this for 11 weeks or get things out. It is still going to change a lot in the next week so if you do grab this version, please test and give me feed back ASAP, inculuding questions on how to do things with the library. This will prompt me to write documentation so I don't have to answer the same question again :-). This 'pre' release version is for people who are interested in the library. The applications will have to be changed to use the new version of the SSL interface. I intend to finish more documentation before I leave but until then, look at the programs in the apps directory. As far as code goes, it is much much nicer than the old version. The current library works, has no memory leaks (as far as I can tell) and is far more bug free that 0.4.5d. There are no global variable of consequence (I believe) and I will produce some documentation that tell where to look for those people that do want to do multi-threaded stuff. There should be more documentation. Have a look in the doc directory. I'll be adding more before I leave, it is a start by mostly documents the crypto library. Tim Hudson will update the web page ASAP. The spelling and grammar are crap but it is better than nothing :-) Reasons to start playing with version 0.5.0 - All the programs in the apps directory build into one ssleay binary. - There is a new version of the 'req' program that generates certificate requests, there is even documentation for this one :-) - There is a demo certification authorithy program. Currently it will look at the simple database and update it. It will generate CRL from the data base. You need to edit the database by hand to revoke a certificate, it is my aim to use perl5/Tk but I don't have time to do this right now. It will generate the certificates but the management scripts still need to be written. This is not a hard task. - Things have been cleaned up alot. - Have a look at the enc and dgst programs in the apps directory. - It supports v3 of x509 certiticates. Major things missing. - I have been working on (and thinging about) the distributed x509 hierachy problem. I have not had time to put my solution in place. It will have to wait until I come back. - I have not put in CRL checking in the certificate verification but it would not be hard to do. I was waiting until I could generate my own CRL (which has only been in the last week) and I don't have time to put it in correctly. - Montgomery multiplication need to be implemented. I know the algorithm, just ran out of time. - PKCS#7. I can load and write the DER version. I need to re-work things to support BER (if that means nothing, read the ASN1 spec :-). - Testing of the higher level digital envelope routines. I have not played with the *_seal() and *_open() type functions. They are written but need testing. The *_sign() and *_verify() functions are rock solid. - PEM. Doing this and PKCS#7 have been dependant on the distributed x509 heirachy problem. I started implementing my ideas, got distracted writing a CA program and then ran out of time. I provide the functionality of RSAref at least. - Re work the asm. code for the x86. I've changed by low level bignum interface again, so I really need to tweak the x86 stuff. gcc is good enough for the other boxes.