mirror of
https://github.com/sqlmapproject/sqlmap.git
synced 2024-11-26 03:23:48 +03:00
78 lines
2.4 KiB
Python
78 lines
2.4 KiB
Python
#!/usr/bin/env python
|
|
|
|
"""
|
|
$Id$
|
|
|
|
Copyright (c) 2006-2010 sqlmap developers (http://sqlmap.sourceforge.net/)
|
|
See the file 'doc/COPYING' for copying permission
|
|
"""
|
|
|
|
from lib.core.exception import sqlmapSyntaxException
|
|
|
|
from plugins.generic.syntax import Syntax as GenericSyntax
|
|
|
|
class Syntax(GenericSyntax):
|
|
def __init__(self):
|
|
GenericSyntax.__init__(self)
|
|
|
|
# As ASCII_CHAR is only available from v2.1 we'll need to adapt this one to use the
|
|
# commented-out part only if detected version>=2.1
|
|
# Reference: wiki.firebirdsql.org/wiki/index.php?page=ASCII_CHAR
|
|
|
|
@staticmethod
|
|
def unescape(expression, quote=True):
|
|
#if quote:
|
|
#while True:
|
|
#index = expression.find("'")
|
|
#if index == -1:
|
|
#break
|
|
|
|
#firstIndex = index + 1
|
|
#index = expression[firstIndex:].find("'")
|
|
|
|
#if index == -1:
|
|
#raise sqlmapSyntaxException, "Unenclosed ' in '%s'" % expression
|
|
|
|
#lastIndex = firstIndex + index
|
|
#old = "'%s'" % expression[firstIndex:lastIndex]
|
|
#unescaped = ""
|
|
|
|
#for i in range(firstIndex, lastIndex):
|
|
#unescaped += "ASCII_CHAR(%d)" % (ord(expression[i]))
|
|
#if i < lastIndex - 1:
|
|
#unescaped += "||"
|
|
|
|
#expression = expression.replace(old, unescaped)
|
|
#else:
|
|
#unescaped = "".join("ASCII_CHAR(%d)||" % ord(c) for c in expression)
|
|
#if unescaped[-1] == "||":
|
|
#unescaped = unescaped[:-1]
|
|
|
|
#expression = unescaped
|
|
|
|
return expression
|
|
|
|
@staticmethod
|
|
def escape(expression):
|
|
#while True:
|
|
#index = expression.find("ASCII_CHAR(")
|
|
#if index == -1:
|
|
#break
|
|
|
|
#firstIndex = index
|
|
#index = expression[firstIndex:].find(")")
|
|
|
|
#if index == -1:
|
|
#raise sqlmapSyntaxException, "Unenclosed ) in '%s'" % expression
|
|
|
|
#lastIndex = firstIndex + index + 1
|
|
#old = expression[firstIndex:lastIndex]
|
|
#oldUpper = old.upper()
|
|
#oldUpper = oldUpper.lstrip("ASCII_CHAR(").rstrip(")")
|
|
#oldUpper = oldUpper.split("||")
|
|
|
|
#escaped = "'%s'" % "".join([chr(int(char)) for char in oldUpper])
|
|
#expression = expression.replace(old, escaped).replace("'||'", "")
|
|
|
|
return expression
|