summaryrefslogtreecommitdiffstats
path: root/crypto/libdes/README
diff options
context:
space:
mode:
Diffstat (limited to 'crypto/libdes/README')
-rw-r--r--crypto/libdes/README54
1 files changed, 54 insertions, 0 deletions
diff --git a/crypto/libdes/README b/crypto/libdes/README
new file mode 100644
index 0000000..58280c2
--- /dev/null
+++ b/crypto/libdes/README
@@ -0,0 +1,54 @@
+
+ libdes, Version 4.01 13-Jan-97
+
+ Copyright (c) 1997, Eric Young
+ All rights reserved.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms specified in COPYRIGHT.
+
+--
+The primary ftp site for this library is
+ftp://ftp.psy.uq.oz.au/pub/Crypto/DES/libdes-x.xx.tar.gz
+libdes is now also shipped with SSLeay. Primary ftp site of
+ftp://ftp.psy.uq.oz.au/pub/Crypto/SSL/SSLeay-x.x.x.tar.gz
+
+The best way to build this library is to build it as part of SSLeay.
+
+This kit builds a DES encryption library and a DES encryption program.
+It supports ecb, cbc, ofb, cfb, triple ecb, triple cbc, triple ofb,
+triple cfb, desx, and MIT's pcbc encryption modes and also has a fast
+implementation of crypt(3).
+It contains support routines to read keys from a terminal,
+generate a random key, generate a key from an arbitrary length string,
+read/write encrypted data from/to a file descriptor.
+
+The implementation was written so as to conform with the manual entry
+for the des_crypt(3) library routines from MIT's project Athena.
+
+destest should be run after compilation to test the des routines.
+rpw should be run after compilation to test the read password routines.
+The des program is a replacement for the sun des command. I believe it
+conforms to the sun version.
+
+The Imakefile is setup for use in the kerberos distribution.
+
+These routines are best compiled with gcc or any other good
+optimising compiler.
+Just turn you optimiser up to the highest settings and run destest
+after the build to make sure everything works.
+
+I believe these routines are close to the fastest and most portable DES
+routines that use small lookup tables (4.5k) that are publicly available.
+The fcrypt routine is faster than ufc's fcrypt (when compiling with
+gcc2 -O2) on the sparc 2 (1410 vs 1270) but is not so good on other machines
+(on a sun3/260 168 vs 336). It is a function of CPU on chip cache size.
+[ 10-Jan-97 and a function of an incorrect speed testing program in
+ ufc which gave much better test figures that reality ].
+
+It is worth noting that on sparc and Alpha CPUs, performance of the DES
+library can vary by upto %10 due to the positioning of files after application
+linkage.
+
+Eric Young (eay@mincom.oz.au)
+
OpenPOWER on IntegriCloud