pferd/PFERD/crawl/ilias/kit_ilias_web_crawler.py
2023-12-09 23:08:42 +01:00

1142 lines
45 KiB
Python

import asyncio
import base64
import os
import re
from collections.abc import Awaitable, Coroutine
from pathlib import PurePath
from typing import Any, Callable, Dict, List, Literal, Optional, Set, Union, cast
from urllib.parse import urljoin
import aiohttp
import yarl
from aiohttp import hdrs
from bs4 import BeautifulSoup, Tag
from ...auth import Authenticator, TfaAuthenticator
from ...config import Config
from ...logging import ProgressBar, log
from ...output_dir import FileSink, Redownload
from ...utils import fmt_path, soupify, url_set_query_param
from ..crawler import AWrapped, CrawlError, CrawlToken, CrawlWarning, DownloadToken, anoncritical
from ..http_crawler import HttpCrawler, HttpCrawlerSection
from .file_templates import Links, learning_module_template
from .ilias_html_cleaner import clean, insert_base_markup
from .kit_ilias_html import (IliasElementType, IliasForumThread, IliasLearningModulePage, IliasPage,
IliasPageElement, _sanitize_path_name, parse_ilias_forum_export)
TargetType = Union[str, int]
_ILIAS_URL = "https://ilias.studium.kit.edu"
class KitShibbolethBackgroundLoginSuccessful():
pass
class KitIliasWebCrawlerSection(HttpCrawlerSection):
def target(self) -> TargetType:
target = self.s.get("target")
if not target:
self.missing_value("target")
if re.fullmatch(r"\d+", target):
# Course id
return int(target)
if target == "desktop":
# Full personal desktop
return target
if target.startswith(_ILIAS_URL):
# ILIAS URL
return target
self.invalid_value("target", target, "Should be <course id | desktop | kit ilias URL>")
def tfa_auth(self, authenticators: Dict[str, Authenticator]) -> Optional[Authenticator]:
value: Optional[str] = self.s.get("tfa_auth")
if value is None:
return None
auth = authenticators.get(value)
if auth is None:
self.invalid_value("tfa_auth", value, "No such auth section exists")
return auth
def links(self) -> Links:
type_str: Optional[str] = self.s.get("links")
if type_str is None:
return Links.FANCY
try:
return Links.from_string(type_str)
except ValueError as e:
self.invalid_value("links", type_str, str(e).capitalize())
def link_redirect_delay(self) -> int:
return self.s.getint("link_redirect_delay", fallback=-1)
def videos(self) -> bool:
return self.s.getboolean("videos", fallback=False)
def forums(self) -> bool:
return self.s.getboolean("forums", fallback=False)
_DIRECTORY_PAGES: Set[IliasElementType] = {
IliasElementType.EXERCISE,
IliasElementType.EXERCISE_FILES,
IliasElementType.FOLDER,
IliasElementType.INFO_TAB,
IliasElementType.MEETING,
IliasElementType.MEDIACAST_VIDEO_FOLDER,
IliasElementType.OPENCAST_VIDEO_FOLDER,
IliasElementType.OPENCAST_VIDEO_FOLDER_MAYBE_PAGINATED,
}
_VIDEO_ELEMENTS: Set[IliasElementType] = {
IliasElementType.MEDIACAST_VIDEO_FOLDER,
IliasElementType.MEDIACAST_VIDEO,
IliasElementType.OPENCAST_VIDEO,
IliasElementType.OPENCAST_VIDEO_PLAYER,
IliasElementType.OPENCAST_VIDEO_FOLDER,
IliasElementType.OPENCAST_VIDEO_FOLDER_MAYBE_PAGINATED,
}
def _iorepeat(attempts: int, name: str, failure_is_error: bool = False) -> Callable[[AWrapped], AWrapped]:
def decorator(f: AWrapped) -> AWrapped:
async def wrapper(*args: Any, **kwargs: Any) -> Optional[Any]:
last_exception: Optional[BaseException] = None
for round in range(attempts):
try:
return await f(*args, **kwargs)
except aiohttp.ContentTypeError: # invalid content type
raise CrawlWarning("ILIAS returned an invalid content type")
except aiohttp.TooManyRedirects:
raise CrawlWarning("Got stuck in a redirect loop")
except aiohttp.ClientPayloadError as e: # encoding or not enough bytes
last_exception = e
except aiohttp.ClientConnectionError as e: # e.g. timeout, disconnect, resolve failed, etc.
last_exception = e
except asyncio.exceptions.TimeoutError as e: # explicit http timeouts in HttpCrawler
last_exception = e
log.explain_topic(f"Retrying operation {name}. Retries left: {attempts - 1 - round}")
if last_exception:
message = f"Error in I/O Operation: {last_exception}"
if failure_is_error:
raise CrawlError(message) from last_exception
else:
raise CrawlWarning(message) from last_exception
raise CrawlError("Impossible return in ilias _iorepeat")
return wrapper # type: ignore
return decorator
def _wrap_io_in_warning(name: str) -> Callable[[AWrapped], AWrapped]:
"""
Wraps any I/O exception in a CrawlWarning.
"""
return _iorepeat(1, name)
def _get_video_cache_key(element: IliasPageElement) -> str:
return f"ilias-video-cache-{element.id()}"
# Crawler control flow:
#
# crawl_desktop -+
# |
# crawl_course --+
# |
# @_io_repeat | # retries internally (before the bar)
# +- crawl_url <-+
# |
# |
# | @_wrap_io_exception # does not need to retry as children acquire bars
# +> crawl_ilias_element -+
# ^ |
# | @_io_repeat | # retries internally (before the bar)
# +- crawl_ilias_page <---+
# | |
# +> get_page | # Handles and retries authentication
# |
# @_io_repeat | # retries internally (before the bar)
# +- download_link <---+
# | |
# +> resolve_target | # Handles and retries authentication
# |
# @_io_repeat | # retries internally (before the bar)
# +- download_video <---+
# | |
# | @_io_repeat | # retries internally (before the bar)
# +- download_file <---+
# |
# +> stream_from_url # Handles and retries authentication
class KitIliasWebCrawler(HttpCrawler):
def __init__(
self,
name: str,
section: KitIliasWebCrawlerSection,
config: Config,
authenticators: Dict[str, Authenticator]
):
# Setting a main authenticator for cookie sharing
auth = section.auth(authenticators)
super().__init__(name, section, config, shared_auth=auth)
if section.tasks() > 1:
log.warn("""
Please avoid using too many parallel requests as these are the KIT ILIAS
instance's greatest bottleneck.
""".strip())
self._shibboleth_login = KitShibbolethLogin(
auth,
section.tfa_auth(authenticators),
)
self._base_url = _ILIAS_URL
self._target = section.target()
self._link_file_redirect_delay = section.link_redirect_delay()
self._links = section.links()
self._videos = section.videos()
self._forums = section.forums()
self._visited_urls: Dict[str, PurePath] = dict()
async def _run(self) -> None:
if isinstance(self._target, int):
log.explain_topic(f"Inferred crawl target: Course with id {self._target}")
await self._crawl_course(self._target)
elif self._target == "desktop":
log.explain_topic("Inferred crawl target: Personal desktop")
await self._crawl_desktop()
else:
log.explain_topic(f"Inferred crawl target: URL {self._target}")
await self._crawl_url(self._target)
async def _crawl_course(self, course_id: int) -> None:
# Start crawling at the given course
root_url = url_set_query_param(
self._base_url + "/goto.php", "target", f"crs_{course_id}"
)
await self._crawl_url(root_url, expected_id=course_id)
async def _crawl_desktop(self) -> None:
appendix = r"ILIAS\PersonalDesktop\PDMainBarProvider|mm_pd_sel_items"
appendix = appendix.encode("ASCII").hex()
await self._crawl_url(self._base_url + "/gs_content.php?item=" + appendix)
async def _crawl_url(self, url: str, expected_id: Optional[int] = None) -> None:
maybe_cl = await self.crawl(PurePath("."))
if not maybe_cl:
return
cl = maybe_cl # Not mypy's fault, but explained here: https://github.com/python/mypy/issues/2608
elements: List[IliasPageElement] = []
# A list as variable redefinitions are not propagated to outer scopes
description: List[BeautifulSoup] = []
@_iorepeat(3, "crawling url")
async def gather_elements() -> None:
elements.clear()
async with cl:
next_stage_url: Optional[str] = url
current_parent = None
# Duplicated code, but the root page is special - we want to avoid fetching it twice!
while next_stage_url:
soup = await self._get_page(next_stage_url, root_page_allowed=True)
if current_parent is None and expected_id is not None:
perma_link_element: Tag = soup.find(id="current_perma_link")
if not perma_link_element or "crs_" not in perma_link_element.get("value"):
raise CrawlError("Invalid course id? Didn't find anything looking like a course")
log.explain_topic(f"Parsing HTML page for {fmt_path(cl.path)}")
log.explain(f"URL: {next_stage_url}")
page = IliasPage(soup, next_stage_url, current_parent)
if next_element := page.get_next_stage_element():
current_parent = next_element
next_stage_url = next_element.url
else:
next_stage_url = None
elements.extend(page.get_child_elements())
if info_tab := page.get_info_tab():
elements.append(info_tab)
if description_string := page.get_description():
description.append(description_string)
# Fill up our task list with the found elements
await gather_elements()
if description:
await self._download_description(PurePath("."), description[0])
elements.sort(key=lambda e: e.id())
tasks: List[Awaitable[None]] = []
for element in elements:
if handle := await self._handle_ilias_element(PurePath("."), element):
tasks.append(asyncio.create_task(handle))
# And execute them
await self.gather(tasks)
async def _handle_ilias_page(
self,
url: str,
parent: IliasPageElement,
path: PurePath,
) -> Optional[Coroutine[Any, Any, None]]:
maybe_cl = await self.crawl(path)
if not maybe_cl:
return None
return self._crawl_ilias_page(url, parent, maybe_cl)
@anoncritical
async def _crawl_ilias_page(
self,
url: str,
parent: IliasPageElement,
cl: CrawlToken,
) -> None:
elements: List[IliasPageElement] = []
# A list as variable redefinitions are not propagated to outer scopes
description: List[BeautifulSoup] = []
@_iorepeat(3, "crawling folder")
async def gather_elements() -> None:
elements.clear()
async with cl:
next_stage_url: Optional[str] = url
current_parent = parent
while next_stage_url:
soup = await self._get_page(next_stage_url)
log.explain_topic(f"Parsing HTML page for {fmt_path(cl.path)}")
log.explain(f"URL: {next_stage_url}")
page = IliasPage(soup, next_stage_url, current_parent)
if next_element := page.get_next_stage_element():
current_parent = next_element
next_stage_url = next_element.url
else:
next_stage_url = None
elements.extend(page.get_child_elements())
if description_string := page.get_description():
description.append(description_string)
# Fill up our task list with the found elements
await gather_elements()
if description:
await self._download_description(cl.path, description[0])
elements.sort(key=lambda e: e.id())
tasks: List[Awaitable[None]] = []
for element in elements:
if handle := await self._handle_ilias_element(cl.path, element):
tasks.append(asyncio.create_task(handle))
# And execute them
await self.gather(tasks)
# These decorators only apply *to this method* and *NOT* to the returned
# awaitables!
# This method does not await the handlers but returns them instead.
# This ensures one level is handled at a time and name deduplication
# works correctly.
@anoncritical
async def _handle_ilias_element(
self,
parent_path: PurePath,
element: IliasPageElement,
) -> Optional[Coroutine[Any, Any, None]]:
if element.url in self._visited_urls:
raise CrawlWarning(
f"Found second path to element {element.name!r} at {element.url!r}. "
+ f"First path: {fmt_path(self._visited_urls[element.url])}. "
+ f"Second path: {fmt_path(parent_path)}."
)
self._visited_urls[element.url] = parent_path
element_path = PurePath(parent_path, element.name)
if element.type in _VIDEO_ELEMENTS:
if not self._videos:
log.status(
"[bold bright_black]",
"Ignored",
fmt_path(element_path),
"[bright_black](enable with option 'videos')"
)
return None
if element.type == IliasElementType.FILE:
return await self._handle_file(element, element_path)
elif element.type == IliasElementType.FORUM:
if not self._forums:
log.status(
"[bold bright_black]",
"Ignored",
fmt_path(element_path),
"[bright_black](enable with option 'forums')"
)
return None
return await self._handle_forum(element, element_path)
elif element.type == IliasElementType.TEST:
log.status(
"[bold bright_black]",
"Ignored",
fmt_path(element_path),
"[bright_black](tests contain no relevant data)"
)
return None
elif element.type == IliasElementType.SURVEY:
log.status(
"[bold bright_black]",
"Ignored",
fmt_path(element_path),
"[bright_black](surveys contain no relevant data)"
)
return None
elif element.type == IliasElementType.SCORM_LEARNING_MODULE:
log.status(
"[bold bright_black]",
"Ignored",
fmt_path(element_path),
"[bright_black](scorm learning modules are not supported)"
)
return None
elif element.type == IliasElementType.LEARNING_MODULE:
return await self._handle_learning_module(element, element_path)
elif element.type == IliasElementType.LINK:
return await self._handle_link(element, element_path)
elif element.type == IliasElementType.BOOKING:
return await self._handle_booking(element, element_path)
elif element.type == IliasElementType.OPENCAST_VIDEO:
return await self._handle_file(element, element_path)
elif element.type == IliasElementType.OPENCAST_VIDEO_PLAYER:
return await self._handle_opencast_video(element, element_path)
elif element.type == IliasElementType.MEDIACAST_VIDEO:
return await self._handle_file(element, element_path)
elif element.type in _DIRECTORY_PAGES:
return await self._handle_ilias_page(element.url, element, element_path)
else:
# This will retry it a few times, failing everytime. It doesn't make any network
# requests, so that's fine.
raise CrawlWarning(f"Unknown element type: {element.type!r}")
async def _handle_link(
self,
element: IliasPageElement,
element_path: PurePath,
) -> Optional[Coroutine[Any, Any, None]]:
log.explain_topic(f"Decision: Crawl Link {fmt_path(element_path)}")
log.explain(f"Links type is {self._links}")
link_template_maybe = self._links.template()
link_extension = self._links.extension()
if not link_template_maybe or not link_extension:
log.explain("Answer: No")
return None
else:
log.explain("Answer: Yes")
element_path = element_path.with_name(element_path.name + link_extension)
maybe_dl = await self.download(element_path, mtime=element.mtime)
if not maybe_dl:
return None
return self._download_link(element, link_template_maybe, maybe_dl)
@anoncritical
@_iorepeat(3, "resolving link")
async def _download_link(self, element: IliasPageElement, link_template: str, dl: DownloadToken) -> None:
async with dl as (bar, sink):
export_url = element.url.replace("cmd=calldirectlink", "cmd=exportHTML")
real_url = await self._resolve_link_target(export_url)
self._write_link_content(link_template, real_url, element.name, element.description, sink)
def _write_link_content(
self,
link_template: str,
url: str,
name: str,
description: Optional[str],
sink: FileSink,
) -> None:
content = link_template
content = content.replace("{{link}}", url)
content = content.replace("{{name}}", name)
content = content.replace("{{description}}", str(description))
content = content.replace("{{redirect_delay}}", str(self._link_file_redirect_delay))
sink.file.write(content.encode("utf-8"))
sink.done()
async def _handle_booking(
self,
element: IliasPageElement,
element_path: PurePath,
) -> Optional[Coroutine[Any, Any, None]]:
log.explain_topic(f"Decision: Crawl Booking Link {fmt_path(element_path)}")
log.explain(f"Links type is {self._links}")
link_template_maybe = self._links.template()
link_extension = self._links.extension()
if not link_template_maybe or not link_extension:
log.explain("Answer: No")
return None
else:
log.explain("Answer: Yes")
element_path = element_path.with_name(element_path.name + link_extension)
maybe_dl = await self.download(element_path, mtime=element.mtime)
if not maybe_dl:
return None
return self._download_booking(element, link_template_maybe, maybe_dl)
@anoncritical
@_iorepeat(1, "downloading description")
async def _download_description(self, parent_path: PurePath, description: BeautifulSoup) -> None:
path = parent_path / "Description.html"
dl = await self.download(path, redownload=Redownload.ALWAYS)
if not dl:
return
async with dl as (bar, sink):
description = clean(insert_base_markup(description))
sink.file.write(description.prettify().encode("utf-8"))
sink.done()
@anoncritical
@_iorepeat(3, "resolving booking")
async def _download_booking(
self,
element: IliasPageElement,
link_template: str,
dl: DownloadToken,
) -> None:
async with dl as (bar, sink):
self._write_link_content(link_template, element.url, element.name, element.description, sink)
async def _resolve_link_target(self, export_url: str) -> str:
async with self.session.get(export_url, allow_redirects=False) as resp:
# No redirect means we were authenticated
if hdrs.LOCATION not in resp.headers:
return soupify(await resp.read()).select_one("a").get("href").strip()
await self._authenticate()
async with self.session.get(export_url, allow_redirects=False) as resp:
# No redirect means we were authenticated
if hdrs.LOCATION not in resp.headers:
return soupify(await resp.read()).select_one("a").get("href").strip()
raise CrawlError("resolve_link_target failed even after authenticating")
async def _handle_opencast_video(
self,
element: IliasPageElement,
element_path: PurePath,
) -> Optional[Coroutine[Any, Any, None]]:
# Copy old mapping as it is likely still relevant
if self.prev_report:
self.report.add_custom_value(
_get_video_cache_key(element),
self.prev_report.get_custom_value(_get_video_cache_key(element))
)
# A video might contain other videos, so let's "crawl" the video first
# to ensure rate limits apply. This must be a download as *this token*
# is re-used if the video consists of a single stream. In that case the
# file name is used and *not* the stream name the ilias html parser reported
# to ensure backwards compatibility.
maybe_dl = await self.download(element_path, mtime=element.mtime, redownload=Redownload.ALWAYS)
# If we do not want to crawl it (user filter), we can move on
if not maybe_dl:
return None
# If we have every file from the cached mapping already, we can ignore this and bail
if self._all_opencast_videos_locally_present(element, maybe_dl.path):
# Mark all existing videos as known to ensure they do not get deleted during cleanup.
# We "downloaded" them, just without actually making a network request as we assumed
# they did not change.
contained = self._previous_contained_opencast_videos(element, maybe_dl.path)
if len(contained) > 1:
# Only do this if we threw away the original dl token,
# to not download single-stream videos twice
for video in contained:
await self.download(video)
return None
return self._download_opencast_video(element, maybe_dl)
def _previous_contained_opencast_videos(
self, element: IliasPageElement, element_path: PurePath
) -> List[PurePath]:
if not self.prev_report:
return []
custom_value = self.prev_report.get_custom_value(_get_video_cache_key(element))
if not custom_value:
return []
cached_value = cast(dict[str, Any], custom_value)
if "known_paths" not in cached_value or "own_path" not in cached_value:
log.explain(f"'known_paths' or 'own_path' missing from cached value: {cached_value}")
return []
transformed_own_path = self._transformer.transform(element_path)
if cached_value["own_path"] != str(transformed_own_path):
log.explain(
f"own_path '{transformed_own_path}' does not match cached value: '{cached_value['own_path']}"
)
return []
return [PurePath(name) for name in cached_value["known_paths"]]
def _all_opencast_videos_locally_present(self, element: IliasPageElement, element_path: PurePath) -> bool:
log.explain_topic(f"Checking local cache for video {fmt_path(element_path)}")
if contained_videos := self._previous_contained_opencast_videos(element, element_path):
log.explain(
f"The following contained videos are known: {','.join(map(fmt_path, contained_videos))}"
)
if all(self._output_dir.resolve(path).exists() for path in contained_videos):
log.explain("Found all known videos locally, skipping enumeration request")
return True
log.explain("Missing at least one video, continuing with requests!")
else:
log.explain("No local cache present")
return False
@anoncritical
@_iorepeat(3, "downloading video")
async def _download_opencast_video(self, element: IliasPageElement, dl: DownloadToken) -> None:
def add_to_report(paths: list[str]) -> None:
self.report.add_custom_value(
_get_video_cache_key(element),
{"known_paths": paths, "own_path": str(self._transformer.transform(dl.path))}
)
async with dl as (bar, sink):
page = IliasPage(await self._get_page(element.url), element.url, element)
stream_elements = page.get_child_elements()
if len(stream_elements) > 1:
log.explain(f"Found multiple video streams for {element.name}")
else:
log.explain(f"Using single video mode for {element.name}")
stream_element = stream_elements[0]
# We do not have a local cache yet
await self._stream_from_url(stream_element.url, sink, bar, is_video=True)
add_to_report([str(self._transformer.transform(dl.path))])
return
contained_video_paths: List[str] = []
for stream_element in stream_elements:
video_path = dl.path.parent / stream_element.name
maybe_dl = await self.download(video_path, mtime=element.mtime, redownload=Redownload.NEVER)
if not maybe_dl:
continue
async with maybe_dl as (bar, sink):
log.explain(f"Streaming video from real url {stream_element.url}")
contained_video_paths.append(str(self._transformer.transform(maybe_dl.path)))
await self._stream_from_url(stream_element.url, sink, bar, is_video=True)
add_to_report(contained_video_paths)
async def _handle_file(
self,
element: IliasPageElement,
element_path: PurePath,
) -> Optional[Coroutine[Any, Any, None]]:
maybe_dl = await self.download(element_path, mtime=element.mtime)
if not maybe_dl:
return None
return self._download_file(element, maybe_dl)
@_iorepeat(3, "downloading file")
@anoncritical
async def _download_file(self, element: IliasPageElement, dl: DownloadToken) -> None:
assert dl # The function is only reached when dl is not None
async with dl as (bar, sink):
await self._stream_from_url(element.url, sink, bar, is_video=False)
async def _stream_from_url(self, url: str, sink: FileSink, bar: ProgressBar, is_video: bool) -> None:
async def try_stream() -> bool:
async with self.session.get(url, allow_redirects=is_video) as resp:
if not is_video:
# Redirect means we weren't authenticated
if hdrs.LOCATION in resp.headers:
return False
# we wanted a video but got HTML
if is_video and "html" in resp.content_type:
return False
if resp.content_length:
bar.set_total(resp.content_length)
async for data in resp.content.iter_chunked(1024):
sink.file.write(data)
bar.advance(len(data))
sink.done()
return True
auth_id = await self._current_auth_id()
if await try_stream():
return
await self.authenticate(auth_id)
if not await try_stream():
raise CrawlError("File streaming failed after authenticate()")
async def _handle_forum(
self,
element: IliasPageElement,
element_path: PurePath,
) -> Optional[Coroutine[Any, Any, None]]:
maybe_cl = await self.crawl(element_path)
if not maybe_cl:
return None
return self._crawl_forum(element, maybe_cl)
@_iorepeat(3, "crawling forum")
@anoncritical
async def _crawl_forum(self, element: IliasPageElement, cl: CrawlToken) -> None:
elements: List[IliasForumThread] = []
async with cl:
next_stage_url = element.url
while next_stage_url:
log.explain_topic(f"Parsing HTML page for {fmt_path(cl.path)}")
log.explain(f"URL: {next_stage_url}")
soup = await self._get_page(next_stage_url)
page = IliasPage(soup, next_stage_url, element)
if next := page.get_next_stage_element():
next_stage_url = next.url
else:
break
download_data = page.get_download_forum_data()
if not download_data:
raise CrawlWarning("Failed to extract forum data")
if download_data.empty:
log.explain("Forum had no threads")
return
html = await self._post_authenticated(download_data.url, download_data.form_data)
elements = parse_ilias_forum_export(soupify(html))
elements.sort(key=lambda elem: elem.title)
tasks: List[Awaitable[None]] = []
for elem in elements:
tasks.append(asyncio.create_task(self._download_forum_thread(cl.path, elem)))
# And execute them
await self.gather(tasks)
@anoncritical
@_iorepeat(3, "saving forum thread")
async def _download_forum_thread(
self,
parent_path: PurePath,
element: IliasForumThread,
) -> None:
path = parent_path / (_sanitize_path_name(element.title) + ".html")
maybe_dl = await self.download(path, mtime=element.mtime)
if not maybe_dl:
return
async with maybe_dl as (bar, sink):
content = element.title_tag.prettify()
content += element.content_tag.prettify()
sink.file.write(content.encode("utf-8"))
sink.done()
async def _handle_learning_module(
self,
element: IliasPageElement,
element_path: PurePath,
) -> Optional[Coroutine[Any, Any, None]]:
maybe_cl = await self.crawl(element_path)
if not maybe_cl:
return None
return self._crawl_learning_module(element, maybe_cl)
@_iorepeat(3, "crawling learning module")
@anoncritical
async def _crawl_learning_module(self, element: IliasPageElement, cl: CrawlToken) -> None:
elements: List[IliasLearningModulePage] = []
async with cl:
log.explain_topic(f"Parsing initial HTML page for {fmt_path(cl.path)}")
log.explain(f"URL: {element.url}")
soup = await self._get_page(element.url)
page = IliasPage(soup, element.url, element)
if next := page.get_learning_module_data():
elements.extend(await self._crawl_learning_module_direction(
cl.path, next.previous_url, "left", element
))
elements.append(next)
elements.extend(await self._crawl_learning_module_direction(
cl.path, next.next_url, "right", element
))
# Reflect their natural ordering in the file names
for index, lm_element in enumerate(elements):
lm_element.title = f"{index:02}_{lm_element.title}"
tasks: List[Awaitable[None]] = []
for index, elem in enumerate(elements):
prev_url = elements[index - 1].title if index > 0 else None
next_url = elements[index + 1].title if index < len(elements) - 1 else None
tasks.append(asyncio.create_task(
self._download_learning_module_page(cl.path, elem, prev_url, next_url)
))
# And execute them
await self.gather(tasks)
async def _crawl_learning_module_direction(
self,
path: PurePath,
start_url: Optional[str],
dir: Union[Literal["left"], Literal["right"]],
parent_element: IliasPageElement
) -> List[IliasLearningModulePage]:
elements: List[IliasLearningModulePage] = []
if not start_url:
return elements
next_element_url: Optional[str] = start_url
counter = 0
while next_element_url:
log.explain_topic(f"Parsing HTML page for {fmt_path(path)} ({dir}-{counter})")
log.explain(f"URL: {next_element_url}")
soup = await self._get_page(next_element_url)
page = IliasPage(soup, next_element_url, parent_element)
if next := page.get_learning_module_data():
elements.append(next)
if dir == "left":
next_element_url = next.previous_url
else:
next_element_url = next.next_url
counter += 1
return elements
@anoncritical
@_iorepeat(3, "saving learning module page")
async def _download_learning_module_page(
self,
parent_path: PurePath,
element: IliasLearningModulePage,
prev: Optional[str],
next: Optional[str]
) -> None:
path = parent_path / (_sanitize_path_name(element.title) + ".html")
maybe_dl = await self.download(path)
if not maybe_dl:
return
my_path = self._transformer.transform(maybe_dl.path)
if not my_path:
return
if prev:
prev_p = self._transformer.transform(parent_path / (_sanitize_path_name(prev) + ".html"))
if prev_p:
prev = os.path.relpath(prev_p, my_path.parent)
else:
prev = None
if next:
next_p = self._transformer.transform(parent_path / (_sanitize_path_name(next) + ".html"))
if next_p:
next = os.path.relpath(next_p, my_path.parent)
else:
next = None
async with maybe_dl as (bar, sink):
content = element.content
content = await self.internalize_images(content)
sink.file.write(learning_module_template(content, maybe_dl.path.name, prev, next).encode("utf-8"))
sink.done()
async def internalize_images(self, tag: Tag) -> Tag:
"""
Tries to fetch ILIAS images and embed them as base64 data.
"""
log.explain_topic("Internalizing images")
for elem in tag.find_all(recursive=True):
if not isinstance(elem, Tag):
continue
if elem.name == "img":
if src := elem.attrs.get("src", None):
url = urljoin(_ILIAS_URL, src)
if not url.startswith(_ILIAS_URL):
continue
log.explain(f"Internalizing {url!r}")
img = await self._get_authenticated(url)
elem.attrs["src"] = "data:;base64," + base64.b64encode(img).decode()
if elem.name == "iframe" and elem.attrs.get("src", "").startswith("//"):
# For unknown reasons the protocol seems to be stripped.
elem.attrs["src"] = "https:" + elem.attrs["src"]
return tag
async def _get_page(self, url: str, root_page_allowed: bool = False) -> BeautifulSoup:
auth_id = await self._current_auth_id()
async with self.session.get(url) as request:
soup = soupify(await request.read())
if IliasPage.is_logged_in(soup):
return self._verify_page(soup, url, root_page_allowed)
# We weren't authenticated, so try to do that
await self.authenticate(auth_id)
# Retry once after authenticating. If this fails, we will die.
async with self.session.get(url) as request:
soup = soupify(await request.read())
if IliasPage.is_logged_in(soup):
return self._verify_page(soup, url, root_page_allowed)
raise CrawlError(f"get_page failed even after authenticating on {url!r}")
@staticmethod
def _verify_page(soup: BeautifulSoup, url: str, root_page_allowed: bool) -> BeautifulSoup:
if IliasPage.is_root_page(soup) and not root_page_allowed:
raise CrawlError(
"Unexpectedly encountered ILIAS root page. "
"This usually happens because the ILIAS instance is broken. "
"If so, wait a day or two and try again. "
"It could also happen because a crawled element links to the ILIAS root page. "
"If so, use a transform with a ! as target to ignore the particular element. "
f"The redirect came from {url}"
)
return soup
async def _post_authenticated(
self,
url: str,
data: dict[str, Union[str, List[str]]]
) -> bytes:
auth_id = await self._current_auth_id()
form_data = aiohttp.FormData()
for key, val in data.items():
form_data.add_field(key, val)
async with self.session.post(url, data=form_data(), allow_redirects=False) as request:
if request.status == 200:
return await request.read()
# We weren't authenticated, so try to do that
await self.authenticate(auth_id)
# Retry once after authenticating. If this fails, we will die.
async with self.session.post(url, data=data, allow_redirects=False) as request:
if request.status == 200:
return await request.read()
raise CrawlError("post_authenticated failed even after authenticating")
async def _get_authenticated(self, url: str) -> bytes:
auth_id = await self._current_auth_id()
async with self.session.get(url, allow_redirects=False) as request:
if request.status == 200:
return await request.read()
# We weren't authenticated, so try to do that
await self.authenticate(auth_id)
# Retry once after authenticating. If this fails, we will die.
async with self.session.get(url, allow_redirects=False) as request:
if request.status == 200:
return await request.read()
raise CrawlError("get_authenticated failed even after authenticating")
# We repeat this as the login method in shibboleth doesn't handle I/O errors.
# Shibboleth is quite reliable as well, the repeat is likely not critical here.
@_iorepeat(3, "Login", failure_is_error=True)
async def _authenticate(self) -> None:
await self._shibboleth_login.login(self.session)
class KitShibbolethLogin:
"""
Login via KIT's shibboleth system.
"""
def __init__(self, authenticator: Authenticator, tfa_authenticator: Optional[Authenticator]) -> None:
self._auth = authenticator
self._tfa_auth = tfa_authenticator
async def login(self, sess: aiohttp.ClientSession) -> None:
"""
Performs the ILIAS Shibboleth authentication dance and saves the login
cookies it receieves.
This function should only be called whenever it is detected that you're
not logged in. The cookies obtained should be good for a few minutes,
maybe even an hour or two.
"""
# Equivalent: Click on "Mit KIT-Account anmelden" button in
# https://ilias.studium.kit.edu/login.php
url = f"{_ILIAS_URL}/shib_login.php"
data = {
"sendLogin": "1",
"idp_selection": "https://idp.scc.kit.edu/idp/shibboleth",
"il_target": "",
"home_organization_selection": "Weiter",
}
soup: Union[BeautifulSoup, KitShibbolethBackgroundLoginSuccessful] = await _shib_post(sess, url, data)
if isinstance(soup, KitShibbolethBackgroundLoginSuccessful):
return
# Attempt to login using credentials, if necessary
while not self._login_successful(soup):
# Searching the form here so that this fails before asking for
# credentials rather than after asking.
form = soup.find("form", {"class": "full content", "method": "post"})
action = form["action"]
csrf_token = form.find("input", {"name": "csrf_token"})["value"]
# Equivalent: Enter credentials in
# https://idp.scc.kit.edu/idp/profile/SAML2/Redirect/SSO
url = "https://idp.scc.kit.edu" + action
username, password = await self._auth.credentials()
data = {
"_eventId_proceed": "",
"j_username": username,
"j_password": password,
"csrf_token": csrf_token
}
soup = await _post(sess, url, data)
if soup.find(id="attributeRelease"):
raise CrawlError(
"ILIAS Shibboleth entitlements changed! "
"Please log in once in your browser and review them"
)
if self._tfa_required(soup):
soup = await self._authenticate_tfa(sess, soup)
if not self._login_successful(soup):
self._auth.invalidate_credentials()
# Equivalent: Being redirected via JS automatically
# (or clicking "Continue" if you have JS disabled)
relay_state = soup.find("input", {"name": "RelayState"})
saml_response = soup.find("input", {"name": "SAMLResponse"})
url = f"{_ILIAS_URL}/Shibboleth.sso/SAML2/POST"
data = { # using the info obtained in the while loop above
"RelayState": relay_state["value"],
"SAMLResponse": saml_response["value"],
}
await sess.post(url, data=data)
async def _authenticate_tfa(
self,
session: aiohttp.ClientSession,
soup: BeautifulSoup
) -> BeautifulSoup:
if not self._tfa_auth:
self._tfa_auth = TfaAuthenticator("ilias-anon-tfa")
tfa_token = await self._tfa_auth.password()
# Searching the form here so that this fails before asking for
# credentials rather than after asking.
form = soup.find("form", {"method": "post"})
action = form["action"]
csrf_token = form.find("input", {"name": "csrf_token"})["value"]
# Equivalent: Enter token in
# https://idp.scc.kit.edu/idp/profile/SAML2/Redirect/SSO
url = "https://idp.scc.kit.edu" + action
data = {
"_eventId_proceed": "",
"j_tokenNumber": tfa_token,
"csrf_token": csrf_token
}
return await _post(session, url, data)
@staticmethod
def _login_successful(soup: BeautifulSoup) -> bool:
relay_state = soup.find("input", {"name": "RelayState"})
saml_response = soup.find("input", {"name": "SAMLResponse"})
return relay_state is not None and saml_response is not None
@staticmethod
def _tfa_required(soup: BeautifulSoup) -> bool:
return soup.find(id="j_tokenNumber") is not None
async def _post(session: aiohttp.ClientSession, url: str, data: Any) -> BeautifulSoup:
async with session.post(url, data=data) as response:
return soupify(await response.read())
async def _shib_post(
session: aiohttp.ClientSession,
url: str,
data: Any
) -> Union[BeautifulSoup, KitShibbolethBackgroundLoginSuccessful]:
"""
aiohttp unescapes '/' and ':' in URL query parameters which is not RFC compliant and rejected
by Shibboleth. Thanks a lot. So now we unroll the requests manually, parse location headers and
build encoded URL objects ourselves... Who thought mangling location header was a good idea??
"""
log.explain_topic("Shib login POST")
async with session.post(url, data=data, allow_redirects=False) as response:
location = response.headers.get("location")
log.explain(f"Got location {location!r}")
if not location:
raise CrawlWarning(f"Login failed (1), no location header present at {url}")
correct_url = yarl.URL(location, encoded=True)
log.explain(f"Corrected location to {correct_url!r}")
if str(correct_url).startswith(_ILIAS_URL):
log.explain("ILIAS recognized our shib token and logged us in in the background, returning")
return KitShibbolethBackgroundLoginSuccessful()
async with session.get(correct_url, allow_redirects=False) as response:
location = response.headers.get("location")
log.explain(f"Redirected to {location!r} with status {response.status}")
# If shib still has a valid session, it will directly respond to the request
if location is None:
log.explain("Shib recognized us, returning its response directly")
return soupify(await response.read())
as_yarl = yarl.URL(response.url)
# Probably not needed anymore, but might catch a few weird situations with a nicer message
if not location or not as_yarl.host:
raise CrawlWarning(f"Login failed (2), no location header present at {correct_url}")
correct_url = yarl.URL.build(
scheme=as_yarl.scheme,
host=as_yarl.host,
path=location,
encoded=True
)
log.explain(f"Corrected location to {correct_url!r}")
async with session.get(correct_url, allow_redirects=False) as response:
return soupify(await response.read())