#!/usr/bin/env python """ $Id$ Copyright (c) 2006-2012 sqlmap developers (http://www.sqlmap.org/) See the file 'doc/COPYING' for copying permission """ import codecs import re import os import StringIO import threading from lib.core.common import Backend from lib.core.common import dataToDumpFile from lib.core.common import dataToStdout from lib.core.common import getUnicode from lib.core.common import normalizeUnicode from lib.core.common import openFile from lib.core.common import restoreDumpMarkedChars from lib.core.common import safeCSValue from lib.core.common import unsafeSQLIdentificatorNaming from lib.core.data import conf from lib.core.data import kb from lib.core.data import logger from lib.core.enums import DBMS from lib.core.exception import sqlmapValueException from lib.core.replication import Replication from lib.core.settings import BUFFERED_LOG_SIZE from lib.core.settings import NULL from lib.core.settings import TRIM_STDOUT_DUMP_SIZE from lib.core.settings import UNICODE_ENCODING class Dump: """ This class defines methods used to parse and output the results of SQL injection actions """ def __init__(self): self._outputFile = None self._outputFP = None self._outputBP = None self._lock = threading.Lock() def _write(self, data, n=True, console=True): text = "%s%s" % (data, "\n" if n else " ") if console: dataToStdout(text) if kb.get("multiThreadMode"): self._lock.acquire() self._outputBP.write(text) if self._outputBP.tell() > BUFFERED_LOG_SIZE: self.flush() if kb.get("multiThreadMode"): self._lock.release() kb.dataOutputFlag = True def flush(self): if self._outputBP and self._outputFP and self._outputBP.tell() > 0: _ = self._outputBP.getvalue() self._outputBP.truncate(0) self._outputFP.write(_) def _formatString(self, inpStr): return restoreDumpMarkedChars(getUnicode(inpStr)) def setOutputFile(self): self._outputFile = "%s%slog" % (conf.outputPath, os.sep) self._outputFP = codecs.open(self._outputFile, "ab", UNICODE_ENCODING) self._outputBP = StringIO.StringIO() def getOutputFile(self): return self._outputFile def string(self, header, data, sort=True): if isinstance(data, (list, tuple, set)): self.lister(header, data, sort) return data = getUnicode(data) if data: data = self._formatString(data) if data[-1] == '\n': data = data[:-1] if "\n" in data: self._write("%s:\n---\n%s\n---\n" % (header, data)) else: self._write("%s: '%s'\n" % (header, data)) else: self._write("%s:\tNone\n" % header) def lister(self, header, elements, sort=True): if elements: self._write("%s [%d]:" % (header, len(elements))) if sort: try: elements = set(elements) elements = list(elements) elements.sort(key=lambda x: x.lower() if isinstance(x, basestring) else x) except: pass for element in elements: if isinstance(element, basestring): self._write("[*] %s" % element) elif isinstance(element, (list, tuple, set)): self._write("[*] " + ", ".join(getUnicode(e) for e in element)) if elements: self._write("") def technic(self, header, data): self.string(header, data) def banner(self,data): self.string("banner", data) def currentUser(self,data): self.string("current user", data) def currentDb(self,data): if Backend.isDbms(DBMS.MAXDB): self.string("current database (no practical usage on %s)" % Backend.getIdentifiedDbms(), data) elif Backend.isDbms(DBMS.ORACLE): self.string("current schema (equivalent to database on %s)" % Backend.getIdentifiedDbms(), data) else: self.string("current database", data) def dba(self,data): self.string("current user is DBA", data) def users(self,users): self.lister("database management system users", users) def userSettings(self, header, userSettings, subHeader): self._areAdmins = set() if userSettings: self._write("%s:" % header) if isinstance(userSettings, (tuple, list, set)): self._areAdmins = userSettings[1] userSettings = userSettings[0] users = userSettings.keys() users.sort(key=lambda x: x.lower() if isinstance(x, basestring) else x) for user in users: settings = userSettings[user] if settings is None: stringSettings = "" else: stringSettings = " [%d]:" % len(settings) if user in self._areAdmins: self._write("[*] %s (administrator)%s" % (user, stringSettings)) else: self._write("[*] %s%s" % (user, stringSettings)) if settings: settings.sort() for setting in settings: self._write(" %s: %s" % (subHeader, setting)) print def dbs(self,dbs): self.lister("available databases", dbs) def dbTables(self, dbTables): if isinstance(dbTables, dict) and len(dbTables) > 0: maxlength = 0 for tables in dbTables.values(): for table in tables: if isinstance(table, (list, tuple, set)): table = table[0] maxlength = max(maxlength, len(normalizeUnicode(table) or str(table))) lines = "-" * (int(maxlength) + 2) for db, tables in dbTables.items(): tables.sort() self._write("Database: %s" % db if db else "Current database") if len(tables) == 1: self._write("[1 table]") else: self._write("[%d tables]" % len(tables)) self._write("+%s+" % lines) for table in tables: if isinstance(table, (list, tuple, set)): table = table[0] blank = " " * (maxlength - len(normalizeUnicode(table) or str(table))) self._write("| %s%s |" % (table, blank)) self._write("+%s+\n" % lines) else: self.string("tables", dbTables) def dbTableColumns(self, tableColumns): if isinstance(tableColumns, dict) and len(tableColumns) > 0: for db, tables in tableColumns.items(): if not db: db = "All" for table, columns in tables.items(): maxlength1 = 0 maxlength2 = 0 colType = None colList = columns.keys() colList.sort(key=lambda x: x.lower() if isinstance(x, basestring) else x) for column in colList: colType = columns[column] maxlength1 = max(maxlength1, len(column or "")) maxlength2 = max(maxlength2, len(colType or "")) maxlength1 = max(maxlength1, len("COLUMN")) lines1 = "-" * (maxlength1 + 2) if colType is not None: maxlength2 = max(maxlength2, len("TYPE")) lines2 = "-" * (maxlength2 + 2) self._write("Database: %s\nTable: %s" % (db if db else "Current database", table)) if len(columns) == 1: self._write("[1 column]") else: self._write("[%d columns]" % len(columns)) if colType is not None: self._write("+%s+%s+" % (lines1, lines2)) else: self._write("+%s+" % lines1) blank1 = " " * (maxlength1 - len("COLUMN")) if colType is not None: blank2 = " " * (maxlength2 - len("TYPE")) if colType is not None: self._write("| Column%s | Type%s |" % (blank1, blank2)) self._write("+%s+%s+" % (lines1, lines2)) else: self._write("| Column%s |" % blank1) self._write("+%s+" % lines1) for column in colList: colType = columns[column] blank1 = " " * (maxlength1 - len(column)) if colType is not None: blank2 = " " * (maxlength2 - len(colType)) self._write("| %s%s | %s%s |" % (column, blank1, colType, blank2)) else: self._write("| %s%s |" % (column, blank1)) if colType is not None: self._write("+%s+%s+\n" % (lines1, lines2)) else: self._write("+%s+\n" % lines1) def dbTablesCount(self, dbTables): if isinstance(dbTables, dict) and len(dbTables) > 0: maxlength1 = len("Table") maxlength2 = len("Entries") for ctables in dbTables.values(): for tables in ctables.values(): for table in tables: maxlength1 = max(maxlength1, len(normalizeUnicode(table) or str(table))) for db, counts in dbTables.items(): self._write("Database: %s" % db if db else "Current database") lines1 = "-" * (maxlength1 + 2) blank1 = " " * (maxlength1 - len("Table")) lines2 = "-" * (maxlength2 + 2) blank2 = " " * (maxlength2 - len("Entries")) self._write("+%s+%s+" % (lines1, lines2)) self._write("| Table%s | Entries%s |" % (blank1, blank2)) self._write("+%s+%s+" % (lines1, lines2)) sortedCounts = counts.keys() sortedCounts.sort(reverse=True) for count in sortedCounts: tables = counts[count] if count is None: count = "Unknown" tables.sort(key=lambda x: x.lower() if isinstance(x, basestring) else x) for table in tables: blank1 = " " * (maxlength1 - len(normalizeUnicode(table) or str(table))) blank2 = " " * (maxlength2 - len(str(count))) self._write("| %s%s | %d%s |" % (table, blank1, count, blank2)) self._write("+%s+%s+\n" % (lines1, lines2)) else: logger.error("unable to retrieve the number of entries for any table") def dbTableValues(self, tableValues): replication = None rtable = None if tableValues is None: return db = tableValues["__infos__"]["db"] if not db: db = "All" table = tableValues["__infos__"]["table"] if conf.replicate: replication = Replication("%s%s%s.sqlite3" % (conf.dumpPath, os.sep, unsafeSQLIdentificatorNaming(db))) else: dumpDbPath = "%s%s%s" % (conf.dumpPath, os.sep, unsafeSQLIdentificatorNaming(db)) if not os.path.isdir(dumpDbPath): os.makedirs(dumpDbPath, 0755) dumpFileName = "%s%s%s.csv" % (dumpDbPath, os.sep, unsafeSQLIdentificatorNaming(table)) dumpFP = openFile(dumpFileName, "wb") count = int(tableValues["__infos__"]["count"]) separator = str() field = 1 fields = len(tableValues) - 1 columns = tableValues.keys() columns.sort(key=lambda x: x.lower() if isinstance(x, basestring) else x) for column in columns: if column != "__infos__": info = tableValues[column] lines = "-" * (int(info["length"]) + 2) separator += "+%s" % lines separator += "+" self._write("Database: %s\nTable: %s" % (db if db else "Current database", table)) if conf.replicate: cols = [] for column in columns: if column != "__infos__": colType = Replication.INTEGER for value in tableValues[column]['values']: try: if not value or re.search("^[\ *]*$", value): #NULL continue int(value) except ValueError: colType = None break if colType is None: colType = Replication.REAL for value in tableValues[column]['values']: try: if not value or re.search("^[\ *]*$", value): #NULL continue float(value) except ValueError: colType = None break cols.append((column, colType if colType else Replication.TEXT)) rtable = replication.createTable(table, cols) if count == 1: self._write("[1 entry]") else: self._write("[%d entries]" % count) self._write(separator) for column in columns: if column != "__infos__": info = tableValues[column] maxlength = int(info["length"]) blank = " " * (maxlength - len(column)) self._write("| %s%s" % (column, blank), n=False) if not conf.replicate: if field == fields: dataToDumpFile(dumpFP, "%s" % safeCSValue(column)) else: dataToDumpFile(dumpFP, "%s%s" % (safeCSValue(column), conf.csvDel)) field += 1 self._write("|\n%s" % separator) if not conf.replicate: dataToDumpFile(dumpFP, "\n") if conf.replicate: rtable.beginTransaction() if count > TRIM_STDOUT_DUMP_SIZE: warnMsg = "console output will be trimmed to " warnMsg += "last %d rows due to " % TRIM_STDOUT_DUMP_SIZE warnMsg += "large table size" logger.warning(warnMsg) for i in xrange(count): console = (i >= count - TRIM_STDOUT_DUMP_SIZE) field = 1 values = [] for column in columns: if column != "__infos__": info = tableValues[column] if len(info["values"]) <= i: continue if info["values"][i] is None: value = u'' else: value = getUnicode(info["values"][i]) if re.search("^[\ *]*$", value): value = NULL values.append(value) maxlength = int(info["length"]) blank = " " * (maxlength - len(value)) self._write("| %s%s" % (value, blank), n=False, console=console) if not conf.replicate: if field == fields: dataToDumpFile(dumpFP, "%s" % safeCSValue(value)) else: dataToDumpFile(dumpFP, "%s%s" % (safeCSValue(value), conf.csvDel)) field += 1 if conf.replicate: try: rtable.insert(values) except sqlmapValueException: pass self._write("|", console=console) if not conf.replicate: dataToDumpFile(dumpFP, "\n") self._write("%s\n" % separator) if conf.replicate: rtable.endTransaction() logger.info("Table '%s.%s' dumped to sqlite3 file '%s'" % (db, table, replication.dbpath)) else: dataToDumpFile(dumpFP, "\n") dumpFP.close() logger.info("Table '%s.%s' dumped to CSV file '%s'" % (db, table, dumpFileName)) def dbColumns(self, dbColumnsDict, colConsider, dbs): for column in dbColumnsDict.keys(): if colConsider == "1": colConsiderStr = "s like '" + column + "' were" else: colConsiderStr = " '%s' was" % column msg = "Column%s found in the " % colConsiderStr msg += "following databases:" self._write(msg) _ = {} for db, tblData in dbs.items(): for tbl, colData in tblData.items(): for col, dataType in colData.items(): if column.lower() in col.lower(): if db in _: if tbl in _[db]: _[db][tbl][col] = dataType else: _[db][tbl] = { col: dataType } else: _[db] = {} _[db][tbl] = { col: dataType } continue self.dbTableColumns(_) def query(self, query, queryRes): self.string(query, queryRes) def rFile(self,filePath,fileData): self.string("%s file saved to" % filePath,fileData,sort=False) def registerValue(self,registerData): self.string("Registry key value data", registerData,sort=False) # object to manage how to print the retrieved queries output to # standard output and sessions file dumper = Dump()