2 Manages the feed cache.
4 @var iface_cache: A singleton cache object. You should normally use this rather than
5 creating new cache objects.
8 # Copyright (C) 2009, Thomas Leonard
9 # See the README file for details, or visit http://0install.net.
13 # We need to know the modification time of each interface, because we refuse
14 # to update to an older version (this prevents an attack where the attacker
15 # sends back an old version which is correctly signed but has a known bug).
17 # The way we store this is a bit complicated due to backward compatibility:
19 # - GPG-signed interfaces have their signatures removed and a last-modified
20 # attribute is stored containing the date from the signature.
22 # - XML-signed interfaces are stored unmodified with their signatures. The
23 # date is extracted from the signature when needed.
25 # - Older versions used to add the last-modified attribute even to files
26 # with XML signatures - these files therefore have invalid signatures and
27 # we extract from the attribute for these.
29 # Eventually, support for the first and third cases will be removed.
32 from logging
import debug
, info
, warn
33 from cStringIO
import StringIO
35 from zeroinstall
import _
36 from zeroinstall
.support
import basedir
37 from zeroinstall
.injector
import reader
, model
38 from zeroinstall
.injector
.namespaces
import config_site
, config_prog
39 from zeroinstall
.injector
.model
import Interface
, escape
, unescape
40 from zeroinstall
import SafeException
42 # If we started a check within this period, don't start another one:
43 FAILED_CHECK_DELAY
= 60 * 60 # 1 Hour
46 assert isinstance(t
, (int, long)), t
47 return time
.strftime('%Y-%m-%d %H:%M:%S UTC', time
.localtime(t
))
49 class ReplayAttack(SafeException
):
50 """Attempt to import a feed that's older than the one in the cache."""
53 class PendingFeed(object):
54 """A feed that has been downloaded but not yet added to the interface cache.
55 Feeds remain in this state until the user confirms that they trust at least
56 one of the signatures.
57 @ivar url: URL for the feed
59 @ivar signed_data: the untrusted data
60 @type signed_data: stream
61 @ivar sigs: signatures extracted from signed_data
62 @type sigs: [L{gpg.Signature}]
63 @ivar new_xml: the payload of the signed_data, or the whole thing if XML
66 __slots__
= ['url', 'signed_data', 'sigs', 'new_xml']
68 def __init__(self
, url
, signed_data
):
69 """Downloaded data is a GPG-signed message.
70 @param url: the URL of the downloaded feed
72 @param signed_data: the downloaded data (not yet trusted)
73 @type signed_data: stream
74 @raise SafeException: if the data is not signed, and logs the actual data"""
76 self
.signed_data
= signed_data
79 def download_keys(self
, fetcher
, feed_hint
= None, key_mirror
= None):
80 """Download any required GPG keys not already on our keyring.
81 When all downloads are done (successful or otherwise), add any new keys
82 to the keyring, L{recheck}.
83 @param fetcher: fetcher to manage the download (was Handler before version 1.5)
84 @type fetcher: L{fetch.Fetcher}
85 @param key_mirror: URL of directory containing keys, or None to use feed's directory
94 key_url
= urlparse
.urljoin(key_mirror
or self
.url
, '%s.gpg' % key_id
)
95 info(_("Fetching key from %s"), key_url
)
96 dl
= fetcher
.download_url(key_url
, hint
= feed_hint
)
97 downloads
[dl
.downloaded
] = (dl
, dl
.tempfile
)
98 blockers
.append(dl
.downloaded
)
103 from zeroinstall
.support
import tasks
111 old_blockers
= blockers
114 for b
in old_blockers
:
118 dl
, stream
= downloads
[b
]
120 self
._downloaded
_key
(stream
)
125 _type
, exception
, tb
= sys
.exc_info()
126 warn(_("Failed to import key for '%(url)s': %(exception)s"), {'url': self
.url
, 'exception': str(exception
)})
128 if exception
and not any_success
:
129 raise exception
, None, tb
133 def _downloaded_key(self
, stream
):
134 import shutil
, tempfile
135 from zeroinstall
.injector
import gpg
137 info(_("Importing key for feed '%s'"), self
.url
)
139 # Python2.4: can't call fileno() on stream, so save to tmp file instead
140 tmpfile
= tempfile
.TemporaryFile(prefix
= 'injector-dl-data-')
142 shutil
.copyfileobj(stream
, tmpfile
)
146 gpg
.import_key(tmpfile
)
151 """Set new_xml and sigs by reading signed_data.
152 You need to call this when previously-missing keys are added to the GPG keyring."""
155 self
.signed_data
.seek(0)
156 stream
, sigs
= gpg
.check_stream(self
.signed_data
)
160 if stream
is not self
.signed_data
:
166 self
.signed_data
.seek(0)
167 info(_("Failed to check GPG signature. Data received was:\n") + repr(self
.signed_data
.read()))
170 class IfaceCache(object):
172 The interface cache stores downloaded and verified interfaces in
173 ~/.cache/0install.net/interfaces (by default).
175 There are methods to query the cache, add to it, check signatures, etc.
177 The cache is updated by L{fetch.Fetcher}.
179 Confusingly, this class is really two caches combined: the in-memory
180 cache of L{model.Interface} objects, and an on-disk cache of L{model.ZeroInstallFeed}s.
181 It will probably be split into two in future.
183 @ivar distro: the native distribution proxy
184 @type distro: L{distro.Distribution}
186 @see: L{iface_cache} - the singleton IfaceCache instance.
189 __slots__
= ['_interfaces', '_feeds', '_distro', '_config']
191 def __init__(self
, distro
= None):
192 """@param distro: distribution used to fetch "distribution:" feeds (since 0.49)
193 @param distro: distribution used to resolve "distribution:" feeds
194 @type distro: L{distro.Distribution}, or None to use the host distribution
196 self
._interfaces
= {}
198 self
._distro
= distro
202 from zeroinstall
.injector
import policy
203 return policy
.get_deprecated_singleton_config().stores
207 if self
._distro
is None:
208 from zeroinstall
.injector
.distro
import get_host_distribution
209 self
._distro
= get_host_distribution()
212 def update_interface_if_trusted(self
, interface
, sigs
, xml
):
214 warnings
.warn("Use update_feed_if_trusted instead", DeprecationWarning, stacklevel
= 2)
215 return self
.update_feed_if_trusted(interface
.uri
, sigs
, xml
)
217 def update_feed_if_trusted(self
, feed_url
, sigs
, xml
):
218 """Update a cached feed (using L{update_feed_from_network})
219 if we trust the signatures.
220 If we don't trust any of the signatures, do nothing.
221 @param feed_url: the feed being updated
223 @param sigs: signatures from L{gpg.check_stream}
224 @type sigs: [L{gpg.Signature}]
225 @param xml: the downloaded replacement feed document
227 @return: True if the feed was updated
232 updated
= self
._oldest
_trusted
(sigs
, trust
.domain_from_url(feed_url
))
233 if updated
is None: return False # None are trusted
235 self
.update_feed_from_network(feed_url
, xml
, updated
)
238 def update_interface_from_network(self
, interface
, new_xml
, modified_time
):
240 warnings
.warn("Use update_feed_from_network instead", DeprecationWarning, stacklevel
= 2)
241 self
.update_feed_from_network(interface
.uri
, new_xml
, modified_time
)
243 def update_feed_from_network(self
, feed_url
, new_xml
, modified_time
):
244 """Update a cached feed.
245 Called by L{update_feed_if_trusted} if we trust this data.
246 After a successful update, L{writer} is used to update the feed's
248 @param feed_url: the feed being updated
249 @type feed_url: L{model.Interface}
250 @param new_xml: the downloaded replacement feed document
252 @param modified_time: the timestamp of the oldest trusted signature
253 (used as an approximation to the feed's modification time)
254 @type modified_time: long
255 @raises ReplayAttack: if modified_time is older than the currently cached time
258 debug(_("Updating '%(interface)s' from network; modified at %(time)s") %
259 {'interface': feed_url
, 'time': _pretty_time(modified_time
)})
261 if '\n<!-- Base64 Signature' not in new_xml
:
262 # Only do this for old-style feeds without
263 # signatures Otherwise, we can get the time from the
264 # signature, and adding this attribute just makes the
266 from xml
.dom
import minidom
267 doc
= minidom
.parseString(new_xml
)
268 doc
.documentElement
.setAttribute('last-modified', str(modified_time
))
270 doc
.writexml(new_xml
)
271 new_xml
= new_xml
.getvalue()
273 self
._import
_new
_feed
(feed_url
, new_xml
, modified_time
)
275 feed
= self
.get_feed(feed_url
)
278 feed
.last_checked
= int(time
.time())
279 writer
.save_feed(feed
)
281 info(_("Updated feed cache entry for %(interface)s (modified %(time)s)"),
282 {'interface': feed
.get_name(), 'time': _pretty_time(modified_time
)})
284 def _import_new_feed(self
, feed_url
, new_xml
, modified_time
):
285 """Write new_xml into the cache.
286 @param feed_url: the URL for the feed being updated
287 @param new_xml: the data to write
288 @param modified_time: when new_xml was modified
289 @raises ReplayAttack: if the new mtime is older than the current one
293 upstream_dir
= basedir
.save_cache_path(config_site
, 'interfaces')
294 cached
= os
.path
.join(upstream_dir
, escape(feed_url
))
297 if os
.path
.exists(cached
):
298 old_xml
= open(cached
).read()
299 if old_xml
== new_xml
:
300 debug(_("No change"))
301 # Update in-memory copy, in case someone else updated the disk copy
302 self
.get_feed(feed_url
, force
= True)
304 old_modified
= int(os
.stat(cached
).st_mtime
)
306 # Do we need to write this temporary file now?
307 stream
= open(cached
+ '.new', 'w')
308 stream
.write(new_xml
)
310 os
.utime(cached
+ '.new', (modified_time
, modified_time
))
311 new_mtime
= reader
.check_readable(feed_url
, cached
+ '.new')
312 assert new_mtime
== modified_time
314 old_modified
= self
._get
_signature
_date
(feed_url
) or old_modified
317 if new_mtime
< old_modified
:
318 os
.unlink(cached
+ '.new')
319 raise ReplayAttack(_("New feed's modification time is "
320 "before old version!\nInterface: %(iface)s\nOld time: %(old_time)s\nNew time: %(new_time)s\n"
322 % {'iface': feed_url
, 'old_time': _pretty_time(old_modified
), 'new_time': _pretty_time(new_mtime
)})
323 if new_mtime
== old_modified
:
324 # You used to have to update the modification time manually.
325 # Now it comes from the signature, this check isn't useful
326 # and often causes problems when the stored format changes
327 # (e.g., when we stopped writing last-modified attributes)
329 #raise SafeException("Interface has changed, but modification time "
330 # "hasn't! Refusing update.")
331 os
.rename(cached
+ '.new', cached
)
332 debug(_("Saved as %s") % cached
)
334 self
.get_feed(feed_url
, force
= True)
336 def get_feed(self
, url
, force
= False, selections_ok
= False):
337 """Get a feed from the cache.
338 @param url: the URL of the feed
339 @param force: load the file from disk again
340 @param selections_ok: if url is a local selections file, return that instead
341 @return: the feed, or None if it isn't cached
342 @rtype: L{model.ZeroInstallFeed}"""
344 feed
= self
._feeds
.get(url
, False)
348 if url
.startswith('distribution:'):
349 master_feed
= self
.get_feed(url
.split(':', 1)[1])
351 return None # Can't happen?
352 feed
= self
.distro
.get_feed(master_feed
)
354 feed
= reader
.load_feed_from_cache(url
, selections_ok
= selections_ok
)
355 if selections_ok
and feed
and not isinstance(feed
, model
.ZeroInstallFeed
):
356 assert feed
.selections
is not None
357 return feed
# (it's actually a selections document)
359 reader
.update_user_feed_overrides(feed
)
360 self
._feeds
[url
] = feed
363 def get_interface(self
, uri
):
364 """Get the interface for uri, creating a new one if required.
365 New interfaces are initialised from the disk cache, but not from
367 @param uri: the URI of the interface to find
368 @rtype: L{model.Interface}
372 assert isinstance(uri
, unicode)
374 if uri
in self
._interfaces
:
375 return self
._interfaces
[uri
]
377 debug(_("Initialising new interface object for %s"), uri
)
378 self
._interfaces
[uri
] = Interface(uri
)
379 reader
.update_from_cache(self
._interfaces
[uri
], iface_cache
= self
)
380 return self
._interfaces
[uri
]
382 def list_all_interfaces(self
):
383 """List all interfaces in the cache.
387 for d
in basedir
.load_cache_paths(config_site
, 'interfaces'):
388 for leaf
in os
.listdir(d
):
389 if not leaf
.startswith('.'):
390 all
.add(unescape(leaf
))
391 return list(all
) # Why not just return the set?
393 def get_icon_path(self
, iface
):
394 """Get the path of a cached icon for an interface.
395 @param iface: interface whose icon we want
396 @return: the path of the cached icon, or None if not cached.
398 return basedir
.load_first_cache(config_site
, 'interface_icons',
401 def get_cached_signatures(self
, uri
):
402 """Verify the cached interface using GPG.
403 Only new-style XML-signed interfaces retain their signatures in the cache.
404 @param uri: the feed to check
406 @return: a list of signatures, or None
407 @rtype: [L{gpg.Signature}] or None
410 if os
.path
.isabs(uri
):
413 old_iface
= basedir
.load_first_cache(config_site
, 'interfaces', escape(uri
))
414 if old_iface
is None:
417 return gpg
.check_stream(open(old_iface
))[1]
418 except SafeException
as ex
:
419 debug(_("No signatures (old-style interface): %s") % ex
)
422 def _get_signature_date(self
, uri
):
423 """Read the date-stamp from the signature of the cached interface.
424 If the date-stamp is unavailable, returns None."""
426 sigs
= self
.get_cached_signatures(uri
)
428 return self
._oldest
_trusted
(sigs
, trust
.domain_from_url(uri
))
430 def _oldest_trusted(self
, sigs
, domain
):
431 """Return the date of the oldest trusted signature in the list, or None if there
432 are no trusted sigs in the list."""
433 trusted
= [s
.get_timestamp() for s
in sigs
if s
.is_trusted(domain
)]
438 def mark_as_checking(self
, url
):
439 """Touch a 'last-check-attempt' timestamp file for this feed.
440 If url is a local path, nothing happens.
441 This prevents us from repeatedly trying to download a failing feed many
442 times in a short period."""
443 if os
.path
.isabs(url
):
445 feeds_dir
= basedir
.save_cache_path(config_site
, config_prog
, 'last-check-attempt')
446 timestamp_path
= os
.path
.join(feeds_dir
, model
._pretty
_escape
(url
))
447 fd
= os
.open(timestamp_path
, os
.O_WRONLY | os
.O_CREAT
, 0o644)
449 os
.utime(timestamp_path
, None) # In case file already exists
451 def get_last_check_attempt(self
, url
):
452 """Return the time of the most recent update attempt for a feed.
453 @see: L{mark_as_checking}
454 @return: The time, or None if none is recorded
455 @rtype: float | None"""
456 timestamp_path
= basedir
.load_first_cache(config_site
, config_prog
, 'last-check-attempt', model
._pretty
_escape
(url
))
458 return os
.stat(timestamp_path
).st_mtime
461 def get_feed_imports(self
, iface
):
462 """Get all feeds that add to this interface.
463 This is the feeds explicitly added by the user, feeds added by the distribution,
464 and feeds imported by a <feed> in the main feed (but not recursively, at present).
467 main_feed
= self
.get_feed(iface
.uri
)
469 return iface
.extra_feeds
+ main_feed
.feeds
471 return iface
.extra_feeds
473 def get_feeds(self
, iface
):
474 """Get all feeds for this interface. This is a mapping from feed URLs
475 to ZeroInstallFeeds. It includes the interface's main feed, plus the
476 resolution of every feed returned by L{get_feed_imports}. Uncached
477 feeds are indicated by a value of None.
478 @rtype: {str: L{ZeroInstallFeed} | None}
480 main_feed
= self
.get_feed(iface
.uri
)
481 results
= {iface
.uri
: main_feed
}
482 for imp
in iface
.extra_feeds
:
484 results
[imp
.uri
] = self
.get_feed(imp
.uri
)
485 except SafeException
as ex
:
486 warn("Failed to load feed '%s: %s", imp
.uri
, ex
)
488 for imp
in main_feed
.feeds
:
489 results
[imp
.uri
] = self
.get_feed(imp
.uri
)
492 def get_implementations(self
, iface
):
493 """Return all implementations from all of iface's feeds.
494 @rtype: [L{Implementation}]
497 for feed
in self
.get_feeds(iface
).itervalues():
499 impls
+= feed
.implementations
.values()
502 def get_feed_targets(self
, feed
):
503 """Return a list of Interfaces for which feed can be a feed.
504 This is used by B{0install add-feed}.
505 @param feed: the feed
506 @type feed: L{model.ZeroInstallFeed} (or, deprecated, a URL)
507 @rtype: [model.Interface]
508 @raise SafeException: If there are no known feeds.
511 if not isinstance(feed
, model
.ZeroInstallFeed
):
513 feed
= self
.get_feed(feed
)
515 raise SafeException("Feed is not cached and using deprecated API")
517 if not feed
.feed_for
:
518 raise SafeException(_("Missing <feed-for> element in '%s'; "
519 "it can't be used as a feed for any other interface.") % feed
.url
)
520 feed_targets
= feed
.feed_for
521 debug(_("Feed targets: %s"), feed_targets
)
522 return [self
.get_interface(uri
) for uri
in feed_targets
]
524 def is_stale(self
, feed
, freshness_threshold
):
525 """Check whether feed needs updating, based on the configured L{config.Config.freshness}.
526 None is considered to be stale.
527 If we already tried to update the feed within FAILED_CHECK_DELAY, returns false.
528 @return: True if feed should be updated
532 if os
.path
.isabs(feed
.url
):
533 return False # Local feeds are never stale
534 if feed
.last_modified
is None:
535 return True # Don't even have it yet
537 staleness
= now
- (feed
.last_checked
or 0)
538 debug(_("Staleness for %(feed)s is %(staleness).2f hours"), {'feed': feed
, 'staleness': staleness
/ 3600.0})
540 if freshness_threshold
<= 0 or staleness
< freshness_threshold
:
541 return False # Fresh enough for us
543 last_check_attempt
= self
.get_last_check_attempt(feed
.url
)
544 if last_check_attempt
and last_check_attempt
> now
- FAILED_CHECK_DELAY
:
545 debug(_("Stale, but tried to check recently (%s) so not rechecking now."), time
.ctime(last_check_attempt
))
550 def usable_feeds(self
, iface
, arch
):
551 """Generator for C{iface.feeds} that are valid for this architecture.
555 for f
in self
.get_feed_imports(iface
):
556 if f
.os
in arch
.os_ranks
and f
.machine
in arch
.machine_ranks
:
559 debug(_("Skipping '%(feed)s'; unsupported architecture %(os)s-%(machine)s"),
560 {'feed': f
, 'os': f
.os
, 'machine': f
.machine
})
562 iface_cache
= IfaceCache()