123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348 |
- import os.path
- import re
- import sqlite3
- import sys
- from random import randint
- def dict_gen(curs: sqlite3.Cursor):
- """
- Generates a dictionary of a sqlite3.Cursor object by fetching the query's results.
- Taken from Python Essential Reference by David Beazley.
- """
- field_names = [d[0] for d in curs.description]
- while True:
- rows = curs.fetchmany()
- if not rows:
- return
- for row in rows:
- yield dict(zip(field_names, row))
- class StatsDatabase:
- def __init__(self, db_path: str):
- """
- Creates a new StatsDatabase.
- :param db_path: The path to the database file
- """
- self.existing_db = os.path.exists(db_path)
- self.database = sqlite3.connect(db_path)
- self.cursor = self.database.cursor()
- # If DB not existing, create a new DB scheme
- if self.existing_db:
- print('Located statistics database at: ', db_path)
- else:
- print('Statistics database not found. Creating new database at: ', db_path)
- def get_file_info(self):
- """
- Retrieves general file statistics from the database. This includes:
- - packetCount : Number of packets in the PCAP file
- - captureDuration : Duration of the packet capture in seconds
- - timestampFirstPacket : Timestamp of the first captured packet
- - timestampLastPacket : Timestamp of the last captured packet
- - avgPacketRate : Average packet rate
- - avgPacketSize : Average packet size
- - avgPacketsSentPerHost : Average number of packets sent per host
- - avgBandwidthIn : Average incoming bandwidth
- - avgBandwidthOut : Average outgoing bandwidth
- :return: a dictionary of keys (see above) and their respective values
- """
- return [r for r in dict_gen(
- self.cursor.execute('SELECT * FROM file_statistics'))][0]
- def get_db_exists(self):
- """
- :return: True if the database was already existent, otherwise False
- """
- return self.existing_db
- @staticmethod
- def _get_selector_keywords():
- """
- :return: a list of selector keywords
- """
- return ['most_used', 'least_used', 'avg', 'all']
- @staticmethod
- def _get_parametrized_selector_keywords():
- """
- :return: a list of parameterizable selector keywords
- """
- return ['ipaddress', 'macaddress']
- @staticmethod
- def _get_extractor_keywords():
- """
- :return: a list of extractor keywords
- """
- return ['random', 'first', 'last']
- def get_all_named_query_keywords(self):
- """
- :return: a list of all named query keywords, used to identify named queries
- """
- return (
- self._get_selector_keywords() + self._get_parametrized_selector_keywords() + self._get_extractor_keywords())
- @staticmethod
- def get_all_sql_query_keywords():
- """
- :return: a list of all supported SQL keywords, used to identify SQL queries
- """
- return ["select", "insert"]
- def _process_user_defined_query(self, query_string: str, query_parameters: tuple = None):
- """
- Takes as input a SQL query query_string and optional a tuple of parameters which are marked by '?' in the query
- and later substituted.
- :param query_string: The query to execute
- :param query_parameters: The tuple of parameters to inject into the query
- :return: the results of the query
- """
- if query_parameters is not None:
- self.cursor.execute(query_string, query_parameters)
- else:
- self.cursor.execute(query_string)
- self.database.commit()
- return self.cursor.fetchall()
- def get_field_types(self, *table_names):
- """
- Creates a dictionary whose keys are the fields of the given table(s) and whose values are the appropriate field
- types, like TEXT for strings and REAL for float numbers.
- :param table_names: The name of table(s)
- :return: a dictionary of {field_name : field_type} for fields of all tables
- """
- dic = {}
- for table in table_names:
- self.cursor.execute("PRAGMA table_info('%s')" % table)
- results = self.cursor.fetchall()
- for field in results:
- dic[field[1].lower()] = field[2]
- return dic
- def named_query_parameterized(self, keyword: str, param_op_val: list):
- """
- Executes a parameterizable named query.
- :param keyword: The query to be executed, like ipaddress or macadress
- :param param_op_val: A list consisting of triples with (parameter, operator, value)
- :return: the results of the executed query
- """
- named_queries = {
- "ipaddress": "SELECT DISTINCT ip_statistics.ipAddress from ip_statistics INNER JOIN ip_mac, ip_ttl, ip_ports, ip_protocols ON ip_statistics.ipAddress=ip_mac.ipAddress AND ip_statistics.ipAddress=ip_ttl.ipAddress AND ip_statistics.ipAddress=ip_ports.ipAddress AND ip_statistics.ipAddress=ip_protocols.ipAddress WHERE ",
- "macaddress": "SELECT DISTINCT macAddress from ip_mac WHERE "}
- query = named_queries.get(keyword)
- field_types = self.get_field_types('ip_mac', 'ip_ttl', 'ip_ports', 'ip_protocols', 'ip_statistics', 'ip_mac')
- conditions = []
- for key, op, value in param_op_val:
- # this makes sure that TEXT fields are queried by strings,
- # e.g. ipAddress=192.168.178.1 --is-converted-to--> ipAddress='192.168.178.1'
- if field_types.get(key) == 'TEXT':
- if not str(value).startswith("'") and not str(value).startswith('"'):
- value = "'" + value + "'"
- # this replacement is required to remove ambiguity in SQL query
- if key == 'ipAddress':
- key = 'ip_mac.ipAddress'
- conditions.append(key + op + str(value))
- where_clause = " AND ".join(conditions)
- query += where_clause
- self.cursor.execute(query)
- return self.cursor.fetchall()
- def _process_named_query(self, query_param_list):
- """
- Executes a named query.
- :param query_param_list: A query list consisting of (keyword, params), e.g. [(most_used, ipAddress), (random,)]
- :return: the result of the query
- """
- # Definition of SQL queries associated to named queries
- named_queries = {
- "most_used.ipaddress": "SELECT ipAddress FROM ip_statistics WHERE (pktsSent+pktsReceived) == (SELECT MAX(pktsSent+pktsReceived) from ip_statistics) LIMIT 1",
- "most_used.macaddress": "SELECT * FROM (SELECT macAddress, COUNT(*) as occ from ip_mac GROUP BY macAddress ORDER BY occ DESC) WHERE occ=(SELECT COUNT(*) as occ from ip_mac GROUP BY macAddress ORDER BY occ DESC LIMIT 1)",
- "most_used.portnumber": "SELECT portNumber, COUNT(portNumber) as cntPort FROM ip_ports GROUP BY portNumber HAVING cntPort=(SELECT MAX(cntPort) from (SELECT portNumber, COUNT(portNumber) as cntPort FROM ip_ports GROUP BY portNumber))",
- "most_used.protocolname": "SELECT protocolName, COUNT(protocolCount) as countProtocols FROM ip_protocols GROUP BY protocolName HAVING countProtocols=(SELECT COUNT(protocolCount) as cnt FROM ip_protocols GROUP BY protocolName ORDER BY cnt DESC LIMIT 1)",
- "most_used.ttlvalue": "SELECT ttlValue FROM ip_ttl GROUP BY ttlValue ORDER BY SUM(ttlCount) DESC LIMIT 1",
- "most_used.mssvalue": "SELECT mssValue FROM tcp_mss GROUP BY mssValue ORDER BY SUM(mssCount) DESC LIMIT 1",
- "most_used.winsize": "SELECT winSize FROM tcp_win GROUP BY winSize ORDER BY SUM(winCount) DESC LIMIT 1",
- "most_used.ipclass": "SELECT ipClass FROM ip_statistics GROUP BY ipClass ORDER BY COUNT(*) DESC LIMIT 1",
- "least_used.ipaddress": "SELECT ipAddress FROM ip_statistics WHERE (pktsSent+pktsReceived) == (SELECT MIN(pktsSent+pktsReceived) from ip_statistics)",
- "least_used.macaddress": "SELECT * FROM (SELECT macAddress, COUNT(*) as occ from ip_mac GROUP BY macAddress ORDER BY occ ASC) WHERE occ=(SELECT COUNT(*) as occ from ip_mac GROUP BY macAddress ORDER BY occ ASC LIMIT 1)",
- "least_used.portnumber": "SELECT portNumber, COUNT(portNumber) as cntPort FROM ip_ports GROUP BY portNumber HAVING cntPort=(SELECT MIN(cntPort) from (SELECT portNumber, COUNT(portNumber) as cntPort FROM ip_ports GROUP BY portNumber))",
- "least_used.protocolname": "SELECT protocolName, COUNT(protocolCount) as countProtocols FROM ip_protocols GROUP BY protocolName HAVING countProtocols=(SELECT COUNT(protocolCount) as cnt FROM ip_protocols GROUP BY protocolName ORDER BY cnt ASC LIMIT 1)",
- "least_used.ttlvalue": "SELECT ttlValue FROM ip_ttl WHERE ttlCount == (SELECT MIN(ttlCount) FROM ip_ttl)",
- "avg.pktsreceived": "SELECT avg(pktsReceived) from ip_statistics",
- "avg.pktssent": "SELECT avg(pktsSent) from ip_statistics",
- "avg.kbytesreceived": "SELECT avg(kbytesReceived) from ip_statistics",
- "avg.kbytessent": "SELECT avg(kbytesSent) from ip_statistics",
- "avg.ttlvalue": "SELECT avg(ttlValue) from ip_ttl",
- "avg.mss": "SELECT avg(mssValue) from tcp_mss",
- "all.ipaddress": "SELECT ipAddress from ip_statistics",
- "all.ttlvalue": "SELECT DISTINCT ttlValue from ip_ttl",
- "all.mss": "SELECT DISTINCT mssValue from tcp_mss",
- "all.macaddress": "SELECT DISTINCT macAddress from ip_mac",
- "all.portnumber": "SELECT DISTINCT portNumber from ip_ports",
- "all.protocolname": "SELECT DISTINCT protocolName from ip_protocols"}
- # Retrieve values by selectors, if given, reduce results by extractor
- last_result = 0
- for q in query_param_list:
- # if selector, like avg, ttl, is given
- if any(e in q[0] for e in self._get_selector_keywords()):
- (keyword, param) = q
- query = named_queries.get(keyword + "." + param)
- self.cursor.execute(str(query))
- last_result = self.cursor.fetchall()
- # if selector is parametrized, i.e. ipAddress(mac=AA:BB:CC:DD:EE) or macAddress(ipAddress=192.168.178.1)
- elif any(e in q[0] for e in self._get_parametrized_selector_keywords()) and any(
- o in q[1] for o in ["<", "=", ">", "<=", ">="]):
- (keyword, param) = q
- # convert string 'paramName1<operator1>paramValue1,paramName2<operator2>paramValue2,...' into list of triples
- param_op_val = [(key, op, value) for (key, op, value) in
- [re.split("(<=|>=|>|<|=)", x) for x in param.split(",")]]
- last_result = self.named_query_parameterized(keyword, param_op_val)
- # if extractor, like random, first, last, is given
- elif any(e in q[0] for e in self._get_extractor_keywords()) and (
- isinstance(last_result, list) or isinstance(last_result, tuple)):
- extractor = q[0]
- if extractor == 'random':
- index = randint(a=0, b=len(last_result) - 1)
- last_result = last_result[index]
- elif extractor == 'first':
- last_result = last_result[0]
- elif extractor == 'last':
- last_result = last_result[-1]
- return last_result
- def process_db_query(self, query_string_in: str, print_results=False, sql_query_parameters: tuple = None):
- """
- Processes a database query. This can either be a standard SQL query or a named query (predefined query).
- :param query_string_in: The string containing the query
- :param print_results: Indicated whether the results should be printed to terminal (True) or not (False)
- :param sql_query_parameters: Parameters for the SQL query (optional)
- :return: the results of the query
- """
- named_query_keywords = self.get_all_named_query_keywords()
- # Clean query_string
- query_string = query_string_in.lower().lstrip()
- # query_string is a user-defined SQL query
- result = None
- if sql_query_parameters is not None or query_string.startswith("select") or query_string.startswith("insert"):
- result = self._process_user_defined_query(query_string, sql_query_parameters)
- # query string is a named query -> parse it and pass it to statisticsDB
- elif any(k in query_string for k in named_query_keywords) and all(k in query_string for k in ['(', ')']):
- # Clean query_string
- query_string = query_string.replace(" ", "")
- # Validity check: Brackets
- brackets_open, brackets_closed = query_string.count("("), query_string.count(")")
- if not (brackets_open == brackets_closed):
- sys.stderr.write("Bracketing of given query '" + query_string + "' is incorrect.")
- # Parse query string into [ (query_keyword1, query_params1), ... ]
- delimiter_start, delimiter_end = "(", ")"
- kplist = []
- current_word = ""
- for char in query_string: # process characters one-by-one
- # if char is no delimiter, add char to current_word
- if char != delimiter_end and char != delimiter_start:
- current_word += char
- # if a start delimiter was found and the current_word so far is a keyword, add it to kplist
- elif char == delimiter_start:
- if current_word in named_query_keywords:
- kplist.append((current_word,))
- current_word = ""
- else:
- print("ERROR: Unrecognized keyword '" + current_word + "' found. Ignoring query.")
- return
- # else if characeter is end delimiter and there were no two directly following ending delimiters,
- # the current_word must be the parameters of an earlier given keyword
- elif char == delimiter_end and len(current_word) > 0:
- kplist[-1] += (current_word,)
- current_word = ""
- result = self._process_named_query(kplist[::-1])
- else:
- sys.stderr.write(
- "Query invalid. Only named queries and SQL SELECT/INSERT allowed. Please check the query's syntax!\n")
- return
- # If result is tuple/list with single element, extract value from list
- requires_extraction = (isinstance(result, list) or isinstance(result, tuple)) and len(result) == 1 and \
- (not isinstance(result[0], tuple) or len(result[0]) == 1)
- while requires_extraction:
- if isinstance(result, list) or isinstance(result, tuple):
- result = result[0]
- else:
- requires_extraction = False
- # If tuple of tuples or list of tuples, each consisting of single element is returned,
- # then convert it into list of values, because the returned colum is clearly specified by the given query
- if (isinstance(result, tuple) or isinstance(result, list)) and all(len(val) == 1 for val in result):
- result = [c for c in result for c in c]
- # Print results if option print_results is True
- if print_results:
- if len(result) == 1 and isinstance(result, list):
- result = result[0]
- print("Query returned 1 record:\n")
- for i in range(0, len(result)):
- print(str(self.cursor.description[i][0]) + ": " + str(result[i]))
- else:
- self._print_query_results(query_string_in, result)
- return result
- def _print_query_results(self, query_string_in: str, result):
- """
- Prints the results of a query.
- Based on http://stackoverflow.com/a/20383011/3017719.
- :param query_string_in: The query the results belong to
- :param result: The results of the query
- """
- # Print number of results according to type of result
- if isinstance(result, list):
- print("Query returned " + str(len(result)) + " records:\n")
- else:
- print("Query returned 1 record:\n")
- # Print query results
- if query_string_in.lstrip().upper().startswith(
- "SELECT") and result is not None and self.cursor.description is not None:
- widths = []
- columns = []
- tavnit = '|'
- separator = '+'
- for cd in self.cursor.description:
- widths.append(len(cd) + 10)
- columns.append(cd[0])
- for w in widths:
- tavnit += " %-" + "%ss |" % (w,)
- separator += '-' * w + '--+'
- print(separator)
- print(tavnit % tuple(columns))
- print(separator)
- if isinstance(result, list):
- for row in result:
- print(tavnit % row)
- else:
- print(tavnit % result)
- print(separator)
- else:
- print(result)
|