• #126 oracle false negative
• #117 #116 - overread in XSS
• #112 fix shared library on macOS
• #122 #115 - false positive issue for XSS
• #113 save space in struct
• #126 add usage to sqli cli tool
• #125 many false positives
• #114 false negative with TSQL and "IF NOT" operation

• Release of whatever changes have been made over the last 2.5 years.

Day-After-Christmas Edition

• No functional changes
• Code reverted to strict C90 style to allow builds on embedded systems, Windows and FreeBSD
• For gcc this means -std=c90 -pedantic, which seems to simulate Windows behavior on Linux
• Other minor style changes to header files.

Black Friday Edition

• Big API Change!! everything in libinjection.h is now libinjection_sqli.h. And a new super simple API is in libinjection.h
• Improvements to folder to prevent bypasses using SQL types (casts). This eliminated about 400 fingerprints as well.
• Blacklisted a very degenerate MySQL ODBC case, that is highly unlike to be used in 'real inputs'. thanks to @LightOS foreporting.. not clear who found it originally.
• Over 400 unit tests now!
• Compiles clean under clang with -Weverything -Wno-padded -Wno-padded is excluded since it's architecture dependant. See clang.sh to see how to invoke.
• PHP documentation fixes, thanks @LightOS

LAMP Special Edition: MySQL and PHP improvements

• Issue #33 Fixes MySQL in latin1-mode use of %A0 as whitespace. This was tricky since %A0 might be part of larger UTF-8 encoding as well. Or perhaps %C2%A0 (utf-8 encoding) might be treated as whitespace. Fortunately, MySQL only seems to treat %A0 as whitespace in latin1 mode. HT @ru_raz0r
• Fixes to Lua testdriver and portability fixes
• Much improved PHP build and test. It now uses phpize and builds and tests like a real module.
• API CHANGE: the macro LIBINJECTION_VERSION has been replaced by const char* libinjection_version(). This allows us to increment the version number without having to regenerate SWIG (or other) bindings for minor releases.

NOTE: Pregenerated SWIG bindings are removed. You'll need to install SWIG before running make. SWIG is packaged on virtually every OS so this should not be a problem.

Here's why:

• Latest versions of swig appear to generate poor quality bindings for LUA and Python. Bugs are filed upstream 1341, 1343, 1345. These are fixed or will be fixed in swig 3.0.0.
• In addition, I've received a number of reports of generated code failing various static analysis
• I can't triangulate which SWIG for which language for which OS will work for you
• I may be switching to libffi for python, and luajit.ffi for lua(jit) in the future, anyways.

• Remove un-needed code

Major Release

• Issue #54: Add test vectors from Arne Swinnen. Thanks qerub@github
• Minor fingerprint update for Issue #54. I don't really think it's valid SQL but it's safe enough to detect without false positives.
• Issue #55: Parse MS SQLSERVER use of [brackets] for column and table names. This is a big one that closes a lot of holes. Thanks nroggle@github
• Issue #56: fix buffer over-read. Thanks safe3@github and flily@github
• Remove use of -fstack-protector as it breaks valgrind detecting memory problems Read more about it http://blog.client9.com/2013/10/12/gcc-valgrind-stackprotector.html
• Fixed folding issue where 1,-sin(1)) would be folded as 1 (1)
• Add more test cases and improved test coverage to 98.8%

• New PHP API
• Big fingerprint update ** about 500 new fingerprints added based on fuzzing tests by Reto Ischi ** about 700 impossible, dead fingerprints removed ** adding folding rule for "sqltype sqltype -> sqltype" since select binary binary binary 1 is valid
• Other minor fingerprints added
• -maybe- API change as typedefs and structs were re-arranged for SWIG

• Fingerprint update -- BETWEEN operation bypasses
• Fingerprint update -- ANY/SOME quasi-function bypasses
• Fixed issue with folding where 1-(2-3) would fold to "nothing" instead of 1
• Improved test coverage to 98.0%
• More adjustments to the PHP/MYSQL backtick to reduce false positives

• Fingerprint update. Credit: Reto Ischi

• found regression in handling of PHP/MySQL backticks. Tests added
• Dead code removed.
• Improved test coverage to 97.7%

• Bug fix for libinjection_sqli_reset @brianrectanus #50
• Non-critical parser fix for numbers with oracle's ending suffix. "SELECT 1FROM .." -> (SELECT, 1, FROM) not (SELECT, 1F, ROM)
• Yet another fix for disambiguating Oracle's "f" suffix for numbers HT @LightOS
• Better parsing of generated number forms of "10.e" and "10.10e" (these are actually table specifiers!) HT @LightOS
• Change sizing of some static arrays to have a length >= 8 For GCC based applications, this allows -fstack-protector to work and -Wstack-protector will now not emit errors.
• Added '-fstack-protector-all -D_FORTIFY_SOURCE=2' to default CFLAGS. About 10% performance loss with -fstack-protector-all
• Improvements in reducing false positives, HT modsecurity team
• Add fingerprint, HT @FluxReiners
• Support for parsing of old ODBC-style typing, e.g. 'select {foo 1};' (valid in MySQL)
• Fix tokenization of "IF EXISTS(....", "IF NOT EXISTS(..."
• Fi possible stack over-read, and improve detection of "sp_password" flag in short sqli HT modsecurity team

• Fingerprint update only HT @LightOS

• Fix regression with COLLATE
• Handle "procedure analyze" under MySQL
• Make API most robust when setting flags
• Add folding API
• Add new all-C test driver to improve testing speed
• Makefile cleanups
• Fired Jenkins! Using in-house system.
• Fixed bypass reported by @FluxReiners

• change how backslash is handled to catch old MSSQL servers sqli See http://websec.ca/kb/sql_injection#MSSQL_Allowed_Intermediary_Chars_AND-OR for details
• Reworking of COLLATE to handle MySQL, TSQL types automatically
• Handle bizarro world TSQL '%1' which is parsed as "0 % 1"
• Better stacked query detection, fixing some regressions
• Folding improvements
• False positive improvements

• Parse binary litterals "0b010101" used by at least mysql and pgsql
• Add fingerprints '1&EUE', '1&EkU' to work around ambiguous parsing rules "-1.for" == '-1.f OR' vs. '-1. FOR' CREDIT @LightOS
• Add parsing rules for COLLATION in MySQL, CREDIT @LightOS
• Reduce false positives by removing all fingerprints that contained "sn"
• Improvement in handling MySQL 'binary' quasi-operator/type
• Improvements in folding
• Removed dependency on SWIG for installing python module

• Fix for parsing Oracle numeric literals
• Fix for oracle whitespace with null char.
• Add unusual SQL join types to keywords lists
• Minor fixes to python API examples

Big Release and Big Engine change. Highly recommend

• Numerous evasions and false positives fixed!
• Tokenizer is now really dumb, and publically exposed. See libinjection_sqli_tokenize.
• Folding engine completely rewritten to be simpler and easier to extend, debug, port.
• MySQL backticks now handled correctly
• @"var" and @'var' parsed correctly (mysql)
• ":=" operator parsed correctly
• non-ascii SQL variables and barewords handled correctly
• less false positives and those that are false positives are more "indeterminate cases" and are only in a few fingerprints
• autogeneration of fingerprints with trivial SQL variations
• support for pgsql $ strings
• support for oracle's q and nq strings
• support for mysql's n strings
• parsing stats exposed
• new swig bindings for python and lua, with callbacks into original scripting language for accept/reject of fingerprints (i.e. manage fingerprints in script, not C code)
• Improved parsing of various special cases in MySQL
• Ban MySQL conditional comments. If we find them, it's marked as SQLi immediately.
• Probably a bunch of other stuff too

All users are advised to upgrade due to risk of DOS

• more fingerprints, more tests
• Issue 34: fix infinite loop

• Add variations on '1U(((', thanks @LightOS
• Add automatically all variations on other cases of 'parens padding'

• Added fingerprint 'nU(kn' and variations, thanks to discussion with @ModSecurity .

• Added fingerprint knknk, thanks @d0znpp

Version 2 is more a software engineering release than SQLi. The API, the code, and filenames are improved for embedded use. Please see the README.md file for details on use.

• Fix Issue30: detection of more small sqli forms with fingerprint "1c".
• Fix Issue32: false positive of '/' of type 'oc' Thanks to @brianrectanus

BIG CHANGES

• File name changes. These are the only relevant files:
  • c/libinjection.h
  • c/libinjection_sqli.c
  • c/libinjection_sqli_data.h
  • COPYING
• Just need to include libinjection.h and link with libinjection_sqli_.c
• sqlparse_private.h and sqli_fingerprints.h are deprecated. Only use #include "libinjection.h"
• API name changes is_sqli and is_string_sqli are now libinjection_is_sqli and libinjection_is_string_sqli
• API change, libinjection_is_sqli now takes a 5th arg for callback data
• API change, libinjection_is_sqli accepts NULL for arg4 and arg5 in which case, a default lookup of fingerprints is used.
• sqlmap_data.json now includes fingerprint information, so people making ports only need to parse one file.

• Allow clang compiler (also in Jenkins, a build with clang and make-scan is done)
• Optimizations should result in > 10% performance improvement for normal workloads
• Add sqlite3 special functions and keywords (since why not)

• fix regression in detecting SQLi of type '1c'

• improved documentation, comments, edits.

• Fix for nested c-style comments used by postgresql and transact-sql. Thanks to @Kanatoko for the report.
• Numerous additions to SQL functions lists (in particular pgsql, transact-sql and ms-access functions) Thanks to Christoffer Sawicki (GitHub "qerub") for report on cut-n-paste error. Thanks to @ryancbarnett for reminder that MS-ACCESS exists ;-)
• Adding of fingerprints to detect HPP attacks.
• Algorihmically added new fingerprints to detect new future sqli attacks. All of these new fingerprints have no been seen 'in the wild' yet.

• Replaced BSD memmem with optimzed version. This eliminates all 3rd party code.
• Added alpha python module (python setup.py install)
• Added sqlparse_fingerprints.h and sqlparse_data.json to aid porting and embeddeding.
• Added version number in sqlparse.h, based on http://www.python.org/dev/peps/pep-0386/#normalizedversion

• retroactive initial release
• all memory issues fixed