# SPDX-FileCopyrightText: 2026 Dennis Fink # # SPDX-License-Identifier: BSD-3-Clause """Fetch and verify incoming Webmentions against their source pages.""" import uuid from datetime import UTC, datetime, timedelta from urllib.parse import urljoin import httpx import rfc3987 import sqlalchemy as sa from bs4 import BeautifulSoup from flask import current_app from .. import APP_NAME, VERSION from .. import DATABASE as db from .. import HUEY as huey from ..models import ReceivedWebmention from ..public_transport import PublicHTTPTransport from ..url_security import AddressResolutionError, UnsafeAddressError RECOVERY_EXHAUSTED_REASON = "Verification recovery limit reached" class VerificationError(Exception): """The source permanently failed ReceivedWebmention verification.""" class SourceGoneError(VerificationError): """The source explicitly reports that it has been removed.""" class TemporaryFetchError(Exception): """Fetching the source may succeed when retried later.""" IRI_PATTERN = rfc3987.get_compiled_pattern("IRI") TRAILING_IRI_PUNCTUATION = ".,;:!?" HTML_URL_ATTRIBUTES = { "href": {"a", "area", "link"}, "src": { "audio", "embed", "iframe", "img", 'input[type="image" i]', "script", "audio source", "video source", "track", "video", }, "cite": {"blockquote", "del", "ins", "q"}, } def html_mentions_target(body: bytes, source_url: str, target_url: str) -> bool: """ Check whether an HTML document mentions the target URL. URL references are resolved against the document's ``base`` element when present, or against the source URL otherwise. :param body: HTML document body. :param source_url: URL from which the document was retrieved. :param target_url: Exact URL to look for. :return: Whether a supported HTML URL attribute references the target URL. """ document = BeautifulSoup(body, "html.parser") base_url = source_url if (base_element := document.select_one("base[href]")) is not None and isinstance( base_href := base_element.get("href"), str ): base_url = urljoin(source_url, base_href.strip()) for attribute, selectors in HTML_URL_ATTRIBUTES.items(): selector = ", ".join( [f"{selector_string}[{attribute}]" for selector_string in selectors] ) for element in document.select(selector): if not isinstance(reference := element.get(attribute), str): continue if urljoin(base_url, reference.strip()) == target_url: return True return False def text_mentions_target(body: str, target_url: str) -> bool: """ Check whether plain text contains the target IRI, ignoring trailing prose punctuation but not extra URL path or query characters. :param body: Plain-text document body. :param target_url: Exact IRI to look for. :return: Whether the target IRI occurs in the document. """ for match in IRI_PATTERN.finditer(body): candidate = match.group() # A URL in prose can be followed by punctuation that the IRI pattern # consumes as part of its match. Only remove trailing punctuation, # leaving longer URLs (e.g. ``target/more``) unmatched. if not candidate.startswith(target_url): continue while candidate != target_url: if candidate[-1] in TRAILING_IRI_PUNCTUATION: candidate = candidate[:-1] elif candidate[-1] == ")" and candidate.count(")") > candidate.count("("): # Keep balanced parentheses that belong to the URL itself. candidate = candidate[:-1] else: break if candidate == target_url: return True return False def fetch_source(source_url: str) -> tuple[httpx.Response, bytes]: """ Fetch a webmention source document. The request follows redirects while enforcing the configured timeout, redirect limit, and maximum response size. :param source_url: URL of the source document. :return: HTTP response and response body. :raises SourceGoneError: If the source returns HTTP 410. :raises TemporaryFetchError: If the source returns a temporary HTTP error. :raises VerificationError: If the source returns another unsuccessful HTTP status or exceeds the configured maximum response size. """ try: return _fetch_source(source_url) except UnsafeAddressError as exc: raise VerificationError("Source resolves to a non-public address") from exc except AddressResolutionError as exc: raise TemporaryFetchError("Source hostname could not be resolved") from exc def _fetch_source(source_url: str) -> tuple[httpx.Response, bytes]: with ( httpx.Client( headers={ "Accept": "text/html, application/xhtml+xml;q=0.9, text/plain;q=0.8", "User-Agent": f"{APP_NAME}/{VERSION} ReceivedWebmention", }, timeout=httpx.Timeout(current_app.config.get("REQUEST_TIMEOUT", 5.0)), follow_redirects=True, max_redirects=current_app.config.get("MAX_REDIRECTS", 20), trust_env=False, transport=PublicHTTPTransport( current_app.config.get("REQUEST_TOTAL_TIMEOUT", 30.0) ), ) as client, client.stream("GET", source_url) as response, ): match response.status_code: case 200: pass case 410: raise SourceGoneError("Source returned HTTP 410") case status: if status in {408, 425, 429} or 500 <= status <= 599: raise TemporaryFetchError(f"Source returned HTTP {status}") else: raise VerificationError(f"Source returned HTTP {status}") max_source_bytes = current_app.config.get("MAX_SOURCE_BYTES", 1_000_000) if (content_length := response.headers.get("Content-Length")) is not None: try: if int(content_length) > max_source_bytes: raise VerificationError("Source document is too large") except ValueError: pass body = bytearray() for chunk in response.iter_bytes(chunk_size=64 * 1024): body.extend(chunk) if len(body) > max_source_bytes: raise VerificationError("Source document is too large") return response, bytes(body) def source_mentions_target(source_url: str, target_url: str) -> bool: """ Fetch a source and check whether it mentions the target URL. HTML and XHTML sources are inspected for supported URL attributes, while plain-text sources are searched for the exact target IRI. :param source_url: URL of the source document. :param target_url: Exact target URL to look for. :return: Whether the source mentions the target URL. :raises VerificationError: If the source has an unsupported or missing content type. :raises SourceGoneError: If the source explicitly reports that it is gone. :raises TemporaryFetchError: If fetching the source fails temporarily. """ response, body = fetch_source(source_url) media_type = ( response.headers.get("Content-Type", "").partition(";")[0].strip().lower() ) match media_type: case "text/html" | "application/xhtml+xml": return html_mentions_target(body, str(response.url), target_url) case "text/plain": try: decoded_body = body.decode( response.encoding or "utf-8", errors="replace" ) except LookupError: decoded_body = body.decode("utf-8", errors="replace") return text_mentions_target(decoded_body, target_url) case _: raise VerificationError( f"Unsupported source content type: {media_type or 'missing'}" ) @huey.task(retries=2, retry_delay=50) def verify_webmention(webmention_uuid: uuid.UUID) -> None: """ Verify a received webmention and persist the result. The transition into ``verifying`` is an atomic claim. Its ``updated_at`` timestamp acts as a lease version: if a reaper has reclaimed the row while the fetch was in progress, the old worker cannot overwrite the new result. Temporary fetch failures are persisted and re-raised for Huey's retries. :param webmention_uuid: Identifier of the received webmention to verify. :raises TemporaryFetchError: If fetching the source fails temporarily. :raises httpx.RequestError: If an HTTP request error occurs while fetching the source. """ webmention = db.session.get(ReceivedWebmention, webmention_uuid) if webmention is None: current_app.logger.warning( "Cannot verify unknown ReceivedWebmention %s", webmention_uuid ) return # Failed tasks may be retried by Huey. Completed and already-running tasks # should not start another concurrent verification. if webmention.status not in {"received", "failed"}: return # A reaper-exhausted row must not be revived by a late Huey retry. if ( webmention.status == "failed" and webmention.failure_reason == RECOVERY_EXHAUSTED_REASON ): return previous_status = webmention.status previous_updated_at = webmention.updated_at source, target = webmention.source, webmention.target started_at = datetime.now(UTC) claimed = db.session.execute( sa.update(ReceivedWebmention) .where( ReceivedWebmention.uuid == webmention_uuid, ReceivedWebmention.status == previous_status, ReceivedWebmention.updated_at == previous_updated_at, ) .values(status="verifying", failure_reason=None, updated_at=started_at) ) if claimed.rowcount != 1: db.session.rollback() return db.session.commit() def finish(status: str, failure_reason: str | None) -> None: # A timeout or a newer verification may have superseded this attempt. # Do not let a late worker overwrite its state. db.session.execute( sa.update(ReceivedWebmention) .where( ReceivedWebmention.uuid == webmention_uuid, ReceivedWebmention.status == "verifying", ReceivedWebmention.updated_at == started_at, ) .values( status=status, failure_reason=failure_reason, updated_at=datetime.now(UTC), ) ) db.session.commit() try: mentions_target = source_mentions_target(source, target) except SourceGoneError as exc: finish("deleted", str(exc)) except VerificationError as exc: finish("failed", str(exc)) except (TemporaryFetchError, httpx.RequestError) as exc: finish("failed", str(exc) or "Source could not be fetched") raise except Exception as exc: # InvalidURL is not a RequestError; other unexpected source errors must # also leave the row in a terminal state, without an endless retry loop. current_app.logger.exception( "Unexpected error verifying ReceivedWebmention %s", webmention_uuid ) finish("failed", str(exc) or type(exc).__name__) else: if mentions_target: finish("verified", None) else: finish("deleted", "Source does not mention target") def recover_stale_webmentions() -> int: """ Requeue verification attempts that were abandoned or lost from the queue. Recover both stale ``verifying`` rows (workers that died) and stale ``received`` rows (tasks that were never enqueued or were lost). Conditional updates prevent overlapping reapers from enqueueing the same row. Each reclaimed row is committed *before* enqueueing, so the worker can claim it. A crash between commit and enqueue is recovered on a later sweep. Persist a recovery count so repeatedly lost or crashing tasks eventually terminate rather than being requeued forever. :return: Number of webmentions requeued. """ now = datetime.now(UTC) cutoff = now - timedelta(seconds=current_app.config["VERIFICATION_TIMEOUT_SECONDS"]) stale = db.session.execute( sa.select( ReceivedWebmention.uuid, ReceivedWebmention.status, ReceivedWebmention.updated_at, ReceivedWebmention.recovery_attempts, ).where( ReceivedWebmention.status.in_(("verifying", "received")), ReceivedWebmention.updated_at <= cutoff, ) ).all() recovered = 0 exhausted = 0 max_recoveries = current_app.config["VERIFICATION_MAX_RECOVERIES"] for identifier, status, updated_at, recovery_attempts in stale: at_limit = recovery_attempts >= max_recoveries changed = db.session.execute( sa.update(ReceivedWebmention) .where( ReceivedWebmention.uuid == identifier, ReceivedWebmention.status == status, ReceivedWebmention.updated_at == updated_at, ReceivedWebmention.recovery_attempts == recovery_attempts, ) .values( status="failed" if at_limit else "received", failure_reason=RECOVERY_EXHAUSTED_REASON if at_limit else None, updated_at=now, recovery_attempts=( recovery_attempts if at_limit else recovery_attempts + 1 ), ) ) if changed.rowcount != 1: db.session.rollback() continue db.session.commit() if at_limit: exhausted += 1 continue verify_webmention(identifier) recovered += 1 if exhausted: current_app.logger.warning( "Failed %d stale webmentions after maximum verification recoveries", exhausted, ) if recovered: current_app.logger.warning( "Requeued %d stale received webmention verifications", recovered ) return recovered