mirror of
https://github.com/carlospolop/PEASS-ng
synced 2024-11-27 14:13:38 +01:00
parser
This commit is contained in:
parent
bc17fb5a3d
commit
83865dddda
@ -1,238 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
import sys
|
||||
import re
|
||||
import json
|
||||
import logging
|
||||
|
||||
# Pattern to identify main section titles
|
||||
MAIN_section_PATTERN = "════════════════════════════════════╣"
|
||||
|
||||
# Main sections (specific for linPEAS)
|
||||
BASIC_INFORMATION_SLUG = "basic-information"
|
||||
SOFTWARE_INFORMATION_SLUG = "software-information"
|
||||
SYSTEM_INFORMATION_SLUG = "system-information"
|
||||
AVAILABLE_SOFTWARE_SLUG = "available-software"
|
||||
NETWORK_INFORMATION_SLUG = "network-information"
|
||||
USERS_INFORMATION_SLUG = "users-information"
|
||||
INTERESTING_FILES_SLUG = "interesting-files"
|
||||
|
||||
try:
|
||||
linpeas_output_path = sys.argv[1]
|
||||
except IndexError as err:
|
||||
# You can pipe the output to "jq" if you have setup
|
||||
print("Error: Needs to pass the .out file\n./linpeas-parser.py <output_file>")
|
||||
sys.exit()
|
||||
|
||||
def basic_information_parser(json_output: dict, row: str) -> dict:
|
||||
"""Returns a dict
|
||||
|
||||
Parses a row a following the boundaries of basic information
|
||||
"""
|
||||
parsed_row = {}
|
||||
|
||||
if ":" in row:
|
||||
parsed_row = {"label": row.replace(":", " ").strip()}
|
||||
elif "[+]" in row:
|
||||
parsed_row = {"label": row.replace("[+]", "").strip()}
|
||||
else:
|
||||
parsed_row = {"label": row}
|
||||
|
||||
return parsed_row
|
||||
|
||||
def software_information_parser(json_output: dict, row: str) -> dict:
|
||||
"""Returns a dict
|
||||
|
||||
Parses a row a following the boundaries of software information
|
||||
"""
|
||||
return {"row": row}
|
||||
|
||||
def system_information_parser(json_output: dict, row: str) -> dict:
|
||||
"""Returns a dict
|
||||
|
||||
Parses a row a following the boundaries of system information
|
||||
"""
|
||||
|
||||
return {"row": row}
|
||||
|
||||
def available_software_parser(json_output: dict, row: str) -> dict:
|
||||
"""Returns a dict
|
||||
|
||||
Parses a row a following the boundaries of available software
|
||||
"""
|
||||
return {"row": row}
|
||||
|
||||
def network_information_parser(json_output: dict, row: str) -> dict:
|
||||
"""Returns a dict
|
||||
|
||||
Parses a row a following the boundaries of network information
|
||||
"""
|
||||
return {"row": row}
|
||||
|
||||
def users_information_parser(json_output: dict, row: str) -> dict:
|
||||
"""Returns a dict
|
||||
|
||||
Parses a row a following the boundaries of network information
|
||||
"""
|
||||
return {"row": row}
|
||||
|
||||
def interesting_files_parser(json_output: dict, row: str) -> dict:
|
||||
"""Returns a dict
|
||||
|
||||
Parses a row a following the boundaries of network information
|
||||
"""
|
||||
return {"row": row}
|
||||
|
||||
def get_parser_by_slug(slug: str):
|
||||
"""Returns a function
|
||||
|
||||
Returns the right parser based on the slug
|
||||
"""
|
||||
return parsers[slug]
|
||||
|
||||
parsers = {
|
||||
BASIC_INFORMATION_SLUG: basic_information_parser,
|
||||
SOFTWARE_INFORMATION_SLUG: software_information_parser,
|
||||
SYSTEM_INFORMATION_SLUG: system_information_parser,
|
||||
AVAILABLE_SOFTWARE_SLUG: available_software_parser,
|
||||
NETWORK_INFORMATION_SLUG: network_information_parser,
|
||||
USERS_INFORMATION_SLUG: users_information_parser,
|
||||
INTERESTING_FILES_SLUG: interesting_files_parser,
|
||||
}
|
||||
|
||||
def read_file(output_path: str) -> [str]:
|
||||
"""Returns a list of strings
|
||||
|
||||
Reads file from a specich path and returns it as a list
|
||||
"""
|
||||
return [row.strip() for row in open(output_path, 'r').readlines() if row]
|
||||
|
||||
def is_starting_section(
|
||||
row: str,
|
||||
pattern: str = MAIN_section_PATTERN
|
||||
) -> bool:
|
||||
"""Returns a boolean
|
||||
|
||||
Checks if row matches the pattern and returns True or False
|
||||
"""
|
||||
return row.find(pattern) > -1
|
||||
|
||||
def extracts_title_label(row: str) -> str:
|
||||
"""Returns a dict
|
||||
|
||||
Extracts a strings whose rows matches the pattern
|
||||
"""
|
||||
return re.findall(r"\w+\s\w+", row)
|
||||
|
||||
def slugify_text(title: str) -> str:
|
||||
"""Returns a dict
|
||||
|
||||
Returns a slugify version of the string.
|
||||
e.g Basic Information -> basic-information
|
||||
"""
|
||||
return title.lower().replace(" ", "-")
|
||||
|
||||
def create_new_main_entry(
|
||||
json_output: object,
|
||||
title: str,
|
||||
row_number: int
|
||||
) -> None:
|
||||
"""Returns None
|
||||
|
||||
Adds a new entry based using "title" as key to return a
|
||||
json output with the initial row number and empty info
|
||||
property where the upcoming information should be added
|
||||
"""
|
||||
|
||||
slug_title = slugify_text(title)
|
||||
|
||||
json_output[slug_title] = {
|
||||
"label": title,
|
||||
"initial_row_number": row_number,
|
||||
"items": {}
|
||||
}
|
||||
|
||||
def get_range_between(
|
||||
json_output: object,
|
||||
section1: str,
|
||||
section2: str
|
||||
) -> list[int, int]:
|
||||
"""Returns a list with two integers
|
||||
|
||||
Extracts the range between one main block and the next one.
|
||||
"""
|
||||
row_number_section1 = json_output[section1]["initial_row_number"] + 1
|
||||
row_number_section2 = json_output[section2]["initial_row_number"] - 1
|
||||
return [row_number_section1, row_number_section2]
|
||||
|
||||
def parse_block(
|
||||
json_output: object,
|
||||
rows: list[str],
|
||||
main_entry_key: str,
|
||||
block_range: list[int, int]
|
||||
) -> None:
|
||||
"""Returns None
|
||||
|
||||
Modifies the "items" from each main section, adding information
|
||||
from the report
|
||||
"""
|
||||
if len(block_range) > 1:
|
||||
initial_row, last_row = block_range
|
||||
row_range = rows[initial_row:last_row]
|
||||
elif len(block_range) == 1:
|
||||
row_range = rows[block_range[0]:]
|
||||
|
||||
slug = slugify_text(main_entry_key)
|
||||
|
||||
items = []
|
||||
|
||||
for row in row_range:
|
||||
sub_section_parser = get_parser_by_slug(slug)
|
||||
items.append(sub_section_parser(json_output, row))
|
||||
|
||||
json_output[main_entry_key]["items"] = items
|
||||
|
||||
def parse_initial_structure(rows: list[str]) -> object:
|
||||
"""Returns an object
|
||||
|
||||
Generates the initial main structure for the json ouput
|
||||
with all the main entries and additional meta properties
|
||||
"""
|
||||
json_output = {}
|
||||
row_number = 0
|
||||
|
||||
for row in rows:
|
||||
if is_starting_section(row, MAIN_section_PATTERN):
|
||||
title = extracts_title_label(row)
|
||||
if len(title) > 0:
|
||||
clean_title = title[0].replace('32m', '')
|
||||
create_new_main_entry(json_output, clean_title, row_number)
|
||||
|
||||
row_number += 1
|
||||
|
||||
return json_output
|
||||
|
||||
def main():
|
||||
rows = read_file(linpeas_output_path)
|
||||
json_output = parse_initial_structure(rows)
|
||||
json_output_keys = list(json_output.keys())
|
||||
keys_length = len(json_output_keys)
|
||||
|
||||
for index in range(0, keys_length):
|
||||
next_index = index + 1
|
||||
if next_index < keys_length:
|
||||
current_label = json_output_keys[index]
|
||||
next_label = json_output_keys[index + 1]
|
||||
|
||||
block_range = get_range_between(json_output, current_label, next_label)
|
||||
parse_block(json_output, rows, current_label, block_range)
|
||||
|
||||
else:
|
||||
last_section_initial_row_number = json_output[next_label]["initial_row_number"] + 1
|
||||
parse_block(json_output, rows, next_label, [last_section_initial_row_number])
|
||||
|
||||
print(json.dumps(json_output))
|
||||
|
||||
if __name__ == "__main__":
|
||||
# execute only if run as a script
|
||||
main()
|
144
parser/peas-parser.py
Executable file
144
parser/peas-parser.py
Executable file
@ -0,0 +1,144 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
import sys
|
||||
import re
|
||||
import json
|
||||
|
||||
# Pattern to identify main section titles
|
||||
TITLE1_PATTERN = r"════════════════════════════════════╣"
|
||||
TITLE2_PATTERN = r"╔══════════╣"
|
||||
TITLE3_PATTERN = r"══╣"
|
||||
INFO_PATTERN = r"╚ "
|
||||
TITLE_CHARS = ['═', '╔', '╣', '╚']
|
||||
|
||||
# Patterns for colors
|
||||
COLORS = {
|
||||
"RED": [r"\x1b\[1;31m"],
|
||||
"GREEN": [r"\x1b\[1;32m"],
|
||||
"YELLOW": [r"\x1b\[1;33m"],
|
||||
"REDYELLOW": [r"\x1b\[1;31;103m"],
|
||||
"BLUE": [r"\x1b\[1;34m"],
|
||||
"LIGHTGREY": [r"\x1b\[1;37m"],
|
||||
"DARKGREY": [r"\x1b\[1;90m"],
|
||||
}
|
||||
|
||||
|
||||
# Final JSON structure
|
||||
FINAL_JSON = {}
|
||||
|
||||
#Constructing the structure
|
||||
C_SECTION = FINAL_JSON
|
||||
C_MAIN_SECTION = FINAL_JSON
|
||||
C_2_SECTION = FINAL_JSON
|
||||
C_3_SECTION = FINAL_JSON
|
||||
|
||||
|
||||
|
||||
|
||||
def is_section(line: str, pattern: str) -> bool:
|
||||
"""Returns a boolean
|
||||
|
||||
Checks if line matches the pattern and returns True or False
|
||||
"""
|
||||
return line.find(pattern) > -1
|
||||
|
||||
def get_colors(line: str) -> dict:
|
||||
"""Given a line return the colored strings"""
|
||||
|
||||
colors = {}
|
||||
for c,regexs in COLORS.items():
|
||||
colors[c] = []
|
||||
for reg in regexs:
|
||||
for re_found in re.findall(reg+".*\x1b", line):
|
||||
colors[c].append(clean_colors(re_found))
|
||||
|
||||
return colors
|
||||
|
||||
def clean_title(line: str) -> str:
|
||||
"""Given a title clean it"""
|
||||
for c in TITLE_CHARS:
|
||||
line = line.replace(c,"")
|
||||
|
||||
line = line.encode("ascii", "ignore").decode() #Remove non ascii chars
|
||||
line = line.strip()
|
||||
return line
|
||||
|
||||
def clean_colors(line: str) -> str:
|
||||
"""Given a line clean the colors inside of it"""
|
||||
|
||||
for reg in re.findall(r'\x1b[^ ]+\dm', line):
|
||||
line = line.replace(reg,"")
|
||||
|
||||
line = line.replace('\x1b',"") #Sometimes that byte stays
|
||||
line = line.strip()
|
||||
return line
|
||||
|
||||
|
||||
def parse_title(line: str) -> str:
|
||||
""" Given a title, clean it"""
|
||||
|
||||
return clean_colors(clean_title(line))
|
||||
|
||||
|
||||
def parse_line(line: str):
|
||||
"""Parse the given line adding it to the FINAL_JSON structure"""
|
||||
|
||||
global FINAL_JSON, C_SECTION, C_MAIN_SECTION, C_2_SECTION, C_3_SECTION
|
||||
|
||||
if is_section(line, TITLE1_PATTERN):
|
||||
title = parse_title(line)
|
||||
FINAL_JSON[title] = { "sections": {}, "lines": [], "infos": [] }
|
||||
C_MAIN_SECTION = FINAL_JSON[title]
|
||||
C_SECTION = C_MAIN_SECTION
|
||||
|
||||
elif is_section(line, TITLE2_PATTERN):
|
||||
title = parse_title(line)
|
||||
FINAL_JSON[C_MAIN_SECTION]["sections"][title] = { "sections": {}, "lines": [], "infos": [] }
|
||||
C_2_SECTION = FINAL_JSON[C_MAIN_SECTION]["sections"][title]
|
||||
C_SECTION = C_2_SECTION
|
||||
|
||||
elif is_section(line, TITLE3_PATTERN):
|
||||
title = parse_title(line)
|
||||
FINAL_JSON[C_MAIN_SECTION]["sections"][C_2_SECTION]["sections"][title] = { "sections": {}, "lines": [], "infos": [] }
|
||||
C_3_SECTION = FINAL_JSON[C_MAIN_SECTION]["sections"][title]
|
||||
C_SECTION = C_3_SECTION
|
||||
|
||||
elif is_section(line, INFO_PATTERN):
|
||||
title = parse_title(line)
|
||||
C_SECTION["infos"].append(title)
|
||||
|
||||
#If here, then it's text
|
||||
else:
|
||||
#If no main section parsed yet, pass
|
||||
if C_SECTION == {}:
|
||||
return
|
||||
|
||||
C_SECTION["lines"].append({
|
||||
"raw_text": line,
|
||||
"clean_text": clean_colors(line),
|
||||
"colors": get_colors(line)
|
||||
})
|
||||
|
||||
|
||||
def main():
|
||||
for line in open(OUTPUT_PATH, 'r').readlines():
|
||||
line = line.strip()
|
||||
if not line:
|
||||
continue
|
||||
|
||||
parse_line(line)
|
||||
|
||||
with open(JSON_PATH, "w") as f:
|
||||
json.dump(FINAL_JSON, f)
|
||||
|
||||
|
||||
# Start execution
|
||||
if __name__ == "__main__":
|
||||
try:
|
||||
OUTPUT_PATH = sys.argv[1]
|
||||
JSON_PATH = sys.argv[2]
|
||||
except IndexError as err:
|
||||
print("Error: Please pass the peas.out file and the path to save the json\n./peas-parser.py <output_file> <json_file.json>")
|
||||
sys.exit(1)
|
||||
|
||||
main()
|
Loading…
Reference in New Issue
Block a user