mirror of
https://github.com/aclist/dztui.git
synced 2026-11-01 07:32:44 +01:00
feat: dzgui 7 alpha
This commit is contained in:
1 parent
1acdbd58c5
commit
2e5b771137
150 files changed
+10359
No files matched your search
@@ -0,0 +1,97 @@
|
||||
import gzip
|
||||
import logging
|
||||
import re
|
||||
import requests
|
||||
import shutil
|
||||
|
||||
from pathlib import Path
|
||||
|
||||
from dzgui.const.constants import REQUEST_TIMEOUT
|
||||
from dzgui.const.endpoints import DB_IP
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
def find_download_url(url: str) -> str | None:
|
||||
res = requests.get(url, timeout=REQUEST_TIMEOUT)
|
||||
res.raise_for_status()
|
||||
reg = r'ref=[\'"]([^\'">]+.csv.gz)'
|
||||
urls = re.findall(reg, res.text)
|
||||
if len(urls) < 1:
|
||||
return None
|
||||
return str(urls[0])
|
||||
|
||||
def get_ipdb(ips_path: Path) -> None:
|
||||
url = find_download_url(DB_IP)
|
||||
if url is None:
|
||||
return
|
||||
|
||||
date = find_date(url)
|
||||
month_file = ips_path.parent / ".month"
|
||||
if month_file.exists() is True:
|
||||
old_date = month_file.read_text().rstrip("\n")
|
||||
if old_date == date:
|
||||
logger.info(f"IP DB date matches: {date}")
|
||||
return
|
||||
|
||||
# TODO: log additional output
|
||||
logger.info(f"Fetching IPDB for {date} from {url}")
|
||||
try:
|
||||
tmp = serialize(url)
|
||||
except Exception as e:
|
||||
logger.critical(e)
|
||||
return
|
||||
|
||||
logger.info(f"Extracting {tmp}")
|
||||
unzip(tmp, ips_path)
|
||||
|
||||
logger.info("Stripping IPv6 records")
|
||||
try:
|
||||
strip_ipv6(ips_path)
|
||||
except Exception as e:
|
||||
logger.critical(e)
|
||||
return
|
||||
|
||||
with open(month_file, "w") as f:
|
||||
f.write(date)
|
||||
logger.info(f"Wrote {date} to {month_file}")
|
||||
|
||||
|
||||
def find_date(url: str) -> str:
|
||||
date = re.sub(r"(dbip-city-lite-)(.*)(.csv.gz)", r"\2", url)
|
||||
date = date.split("/")[-1]
|
||||
return date
|
||||
|
||||
|
||||
def unzip(tmp: Path, ips_path: Path) -> None:
|
||||
with gzip.open(tmp, "rb") as f:
|
||||
with open(ips_path, "wb") as outfile:
|
||||
shutil.copyfileobj(f, outfile)
|
||||
tmp.unlink()
|
||||
|
||||
|
||||
def serialize(url: str) -> Path:
|
||||
with requests.get(url, stream=True) as r:
|
||||
r.raise_for_status()
|
||||
filename = "ips.csv.gz"
|
||||
tmp = Path("/tmp") / filename
|
||||
with open(tmp, "wb") as f:
|
||||
shutil.copyfileobj(r.raw, f)
|
||||
return tmp
|
||||
|
||||
|
||||
# TODO: optimize this function
|
||||
# consider using grep
|
||||
# cf. grep -vE "^[a-z0-9]{4}:" | grep -v "::" > "$ip_file"
|
||||
def strip_ipv6(path: Path) -> None:
|
||||
ips = []
|
||||
with open(path, "r") as f:
|
||||
s = f.read()
|
||||
|
||||
reg = r"^\d{1,3}\..*"
|
||||
ips = re.findall(reg, s, re.MULTILINE)
|
||||
assert ips[0].split(",")[0] == "0.0.0.0"
|
||||
assert ips[-1].split(",")[0] == "224.0.0.0"
|
||||
|
||||
with open(path, "w") as f:
|
||||
for ip in ips:
|
||||
f.write(ip + "\n")
|
||||
Reference in new issue
Block a user