3 from itertools import count
4 from zope.interface import implements
5 from twisted.internet import defer
6 from twisted.python import failure
7 from twisted.internet.interfaces import IPushProducer, IConsumer
8 from foolscap.api import eventually, fireEventually
9 from allmydata.interfaces import IRetrieveStatus, NotEnoughSharesError, \
10 DownloadStopped, MDMF_VERSION, SDMF_VERSION
11 from allmydata.util import hashutil, log, mathutil
12 from allmydata.util.dictutil import DictOfSets
13 from allmydata import hashtree, codec
14 from allmydata.storage.server import si_b2a
15 from pycryptopp.cipher.aes import AES
16 from pycryptopp.publickey import rsa
18 from allmydata.mutable.common import CorruptShareError, UncoordinatedWriteError
19 from allmydata.mutable.layout import MDMFSlotReadProxy
22 implements(IRetrieveStatus)
23 statusid_counter = count(0)
26 self.timings["fetch_per_server"] = {}
27 self.timings["decode"] = 0.0
28 self.timings["decrypt"] = 0.0
29 self.timings["cumulative_verify"] = 0.0
32 self.storage_index = None
34 self.encoding = ("?","?")
36 self.status = "Not started"
38 self.counter = self.statusid_counter.next()
39 self.started = time.time()
41 def get_started(self):
43 def get_storage_index(self):
44 return self.storage_index
45 def get_encoding(self):
47 def using_helper(self):
53 def get_progress(self):
57 def get_counter(self):
60 def add_fetch_timing(self, peerid, elapsed):
61 if peerid not in self.timings["fetch_per_server"]:
62 self.timings["fetch_per_server"][peerid] = []
63 self.timings["fetch_per_server"][peerid].append(elapsed)
64 def accumulate_decode_time(self, elapsed):
65 self.timings["decode"] += elapsed
66 def accumulate_decrypt_time(self, elapsed):
67 self.timings["decrypt"] += elapsed
68 def set_storage_index(self, si):
69 self.storage_index = si
70 def set_helper(self, helper):
72 def set_encoding(self, k, n):
73 self.encoding = (k, n)
74 def set_size(self, size):
76 def set_status(self, status):
78 def set_progress(self, value):
80 def set_active(self, value):
87 # this class is currently single-use. Eventually (in MDMF) we will make
88 # it multi-use, in which case you can call download(range) multiple
89 # times, and each will have a separate response chain. However the
90 # Retrieve object will remain tied to a specific version of the file, and
91 # will use a single ServerMap instance.
92 implements(IPushProducer)
94 def __init__(self, filenode, servermap, verinfo, fetch_privkey=False,
97 assert self._node.get_pubkey()
98 self._storage_index = filenode.get_storage_index()
99 assert self._node.get_readkey()
100 self._last_failure = None
101 prefix = si_b2a(self._storage_index)[:5]
102 self._log_number = log.msg("Retrieve(%s): starting" % prefix)
103 self._outstanding_queries = {} # maps (peerid,shnum) to start_time
105 self._decoding = False
106 self._bad_shares = set()
108 self.servermap = servermap
109 assert self._node.get_pubkey()
110 self.verinfo = verinfo
111 # during repair, we may be called upon to grab the private key, since
112 # it wasn't picked up during a verify=False checker run, and we'll
113 # need it for repair to generate a new version.
114 self._need_privkey = verify or (fetch_privkey
115 and not self._node.get_privkey())
117 if self._need_privkey:
118 # TODO: Evaluate the need for this. We'll use it if we want
119 # to limit how many queries are on the wire for the privkey
121 self._privkey_query_markers = [] # one Marker for each time we've
122 # tried to get the privkey.
124 # verify means that we are using the downloader logic to verify all
125 # of our shares. This tells the downloader a few things.
127 # 1. We need to download all of the shares.
128 # 2. We don't need to decode or decrypt the shares, since our
129 # caller doesn't care about the plaintext, only the
130 # information about which shares are or are not valid.
131 # 3. When we are validating readers, we need to validate the
132 # signature on the prefix. Do we? We already do this in the
134 self._verify = verify
136 self._status = RetrieveStatus()
137 self._status.set_storage_index(self._storage_index)
138 self._status.set_helper(False)
139 self._status.set_progress(0.0)
140 self._status.set_active(True)
141 (seqnum, root_hash, IV, segsize, datalength, k, N, prefix,
142 offsets_tuple) = self.verinfo
143 self._status.set_size(datalength)
144 self._status.set_encoding(k, N)
146 self._stopped = False
147 self._pause_deferred = None
149 self._read_length = None
150 self.log("got seqnum %d" % self.verinfo[0])
153 def get_status(self):
156 def log(self, *args, **kwargs):
157 if "parent" not in kwargs:
158 kwargs["parent"] = self._log_number
159 if "facility" not in kwargs:
160 kwargs["facility"] = "tahoe.mutable.retrieve"
161 return log.msg(*args, **kwargs)
163 def _set_current_status(self, state):
164 seg = "%d/%d" % (self._current_segment, self._last_segment)
165 self._status.set_status("segment %s (%s)" % (seg, state))
170 def pauseProducing(self):
172 I am called by my download target if we have produced too much
173 data for it to handle. I make the downloader stop producing new
174 data until my resumeProducing method is called.
176 if self._pause_deferred is not None:
179 # fired when the download is unpaused.
180 self._old_status = self._status.get_status()
181 self._set_current_status("paused")
183 self._pause_deferred = defer.Deferred()
186 def resumeProducing(self):
188 I am called by my download target once it is ready to begin
189 receiving data again.
191 if self._pause_deferred is None:
194 p = self._pause_deferred
195 self._pause_deferred = None
196 self._status.set_status(self._old_status)
198 eventually(p.callback, None)
200 def stopProducing(self):
202 self.resumeProducing()
205 def _check_for_paused(self, res):
207 I am called just before a write to the consumer. I return a
208 Deferred that eventually fires with the data that is to be
209 written to the consumer. If the download has not been paused,
210 the Deferred fires immediately. Otherwise, the Deferred fires
211 when the downloader is unpaused.
214 raise DownloadStopped("our Consumer called stopProducing()")
215 if self._pause_deferred is not None:
217 self._pause_deferred.addCallback(lambda ignored: d.callback(res))
219 return defer.succeed(res)
222 def download(self, consumer=None, offset=0, size=None):
223 assert IConsumer.providedBy(consumer) or self._verify
226 self._consumer = consumer
227 # we provide IPushProducer, so streaming=True, per
229 self._consumer.registerProducer(self, streaming=True)
231 self._done_deferred = defer.Deferred()
232 self._offset = offset
233 self._read_length = size
234 self._setup_download()
235 self._setup_encoding_parameters()
236 self.log("starting download")
237 self._started_fetching = time.time()
238 # The download process beyond this is a state machine.
239 # _add_active_peers will select the peers that we want to use
240 # for the download, and then attempt to start downloading. After
241 # each segment, it will check for doneness, reacting to broken
242 # peers and corrupt shares as necessary. If it runs out of good
243 # peers before downloading all of the segments, _done_deferred
244 # will errback. Otherwise, it will eventually callback with the
245 # contents of the mutable file.
247 return self._done_deferred
250 d = fireEventually(None) # avoid #237 recursion limit problem
251 d.addCallback(lambda ign: self._activate_enough_peers())
252 d.addCallback(lambda ign: self._download_current_segment())
253 # when we're done, _download_current_segment will call _done. If we
254 # aren't, it will call loop() again.
255 d.addErrback(self._error)
257 def _setup_download(self):
258 self._started = time.time()
259 self._status.set_status("Retrieving Shares")
261 # how many shares do we need?
270 offsets_tuple) = self.verinfo
272 # first, which servers can we use?
273 versionmap = self.servermap.make_versionmap()
274 shares = versionmap[self.verinfo]
275 # this sharemap is consumed as we decide to send requests
276 self.remaining_sharemap = DictOfSets()
277 for (shnum, peerid, timestamp) in shares:
278 self.remaining_sharemap.add(shnum, peerid)
279 # If the servermap update fetched anything, it fetched at least 1
280 # KiB, so we ask for that much.
281 # TODO: Change the cache methods to allow us to fetch all of the
282 # data that they have, then change this method to do that.
283 any_cache = self._node._read_from_cache(self.verinfo, shnum,
285 ss = self.servermap.connections[peerid]
286 reader = MDMFSlotReadProxy(ss,
290 reader.peerid = peerid
291 self.readers[shnum] = reader
292 assert len(self.remaining_sharemap) >= k
294 self.shares = {} # maps shnum to validated blocks
295 self._active_readers = [] # list of active readers for this dl.
296 self._block_hash_trees = {} # shnum => hashtree
298 # We need one share hash tree for the entire file; its leaves
299 # are the roots of the block hash trees for the shares that
300 # comprise it, and its root is in the verinfo.
301 self.share_hash_tree = hashtree.IncompleteHashTree(N)
302 self.share_hash_tree.set_hashes({0: root_hash})
304 def decode(self, blocks_and_salts, segnum):
306 I am a helper method that the mutable file update process uses
307 as a shortcut to decode and decrypt the segments that it needs
308 to fetch in order to perform a file update. I take in a
309 collection of blocks and salts, and pick some of those to make a
310 segment with. I return the plaintext associated with that
313 # shnum => block hash tree. Unused, but setup_encoding_parameters will
315 self._block_hash_trees = None
316 self._setup_encoding_parameters()
318 # This is the form expected by decode.
319 blocks_and_salts = blocks_and_salts.items()
320 blocks_and_salts = [(True, [d]) for d in blocks_and_salts]
322 d = self._decode_blocks(blocks_and_salts, segnum)
323 d.addCallback(self._decrypt_segment)
327 def _setup_encoding_parameters(self):
329 I set up the encoding parameters, including k, n, the number
330 of segments associated with this file, and the segment decoders.
340 offsets_tuple) = self.verinfo
341 self._required_shares = k
342 self._total_shares = n
343 self._segment_size = segsize
344 self._data_length = datalength
347 self._version = MDMF_VERSION
349 self._version = SDMF_VERSION
351 if datalength and segsize:
352 self._num_segments = mathutil.div_ceil(datalength, segsize)
353 self._tail_data_size = datalength % segsize
355 self._num_segments = 0
356 self._tail_data_size = 0
358 self._segment_decoder = codec.CRSDecoder()
359 self._segment_decoder.set_params(segsize, k, n)
361 if not self._tail_data_size:
362 self._tail_data_size = segsize
364 self._tail_segment_size = mathutil.next_multiple(self._tail_data_size,
365 self._required_shares)
366 if self._tail_segment_size == self._segment_size:
367 self._tail_decoder = self._segment_decoder
369 self._tail_decoder = codec.CRSDecoder()
370 self._tail_decoder.set_params(self._tail_segment_size,
371 self._required_shares,
374 self.log("got encoding parameters: "
377 "%d segments of %d bytes each (%d byte tail segment)" % \
378 (k, n, self._num_segments, self._segment_size,
379 self._tail_segment_size))
381 if self._block_hash_trees is not None:
382 for i in xrange(self._total_shares):
383 # So we don't have to do this later.
384 self._block_hash_trees[i] = hashtree.IncompleteHashTree(self._num_segments)
386 # Our last task is to tell the downloader where to start and
387 # where to stop. We use three parameters for that:
388 # - self._start_segment: the segment that we need to start
390 # - self._current_segment: the next segment that we need to
392 # - self._last_segment: The last segment that we were asked to
395 # We say that the download is complete when
396 # self._current_segment > self._last_segment. We use
397 # self._start_segment and self._last_segment to know when to
398 # strip things off of segments, and how much to strip.
400 self.log("got offset: %d" % self._offset)
401 # our start segment is the first segment containing the
402 # offset we were given.
403 start = self._offset // self._segment_size
405 assert start < self._num_segments
406 self._start_segment = start
407 self.log("got start segment: %d" % self._start_segment)
409 self._start_segment = 0
412 # If self._read_length is None, then we want to read the whole
413 # file. Otherwise, we want to read only part of the file, and
414 # need to figure out where to stop reading.
415 if self._read_length is not None:
416 # our end segment is the last segment containing part of the
417 # segment that we were asked to read.
418 self.log("got read length %d" % self._read_length)
419 if self._read_length != 0:
420 end_data = self._offset + self._read_length
422 # We don't actually need to read the byte at end_data,
423 # but the one before it.
424 end = (end_data - 1) // self._segment_size
426 assert end < self._num_segments
427 self._last_segment = end
429 self._last_segment = self._start_segment
430 self.log("got end segment: %d" % self._last_segment)
432 self._last_segment = self._num_segments - 1
434 self._current_segment = self._start_segment
436 def _activate_enough_peers(self):
438 I populate self._active_readers with enough active readers to
439 retrieve the contents of this mutable file. I am called before
440 downloading starts, and (eventually) after each validation
441 error, connection error, or other problem in the download.
443 # TODO: It would be cool to investigate other heuristics for
444 # reader selection. For instance, the cost (in time the user
445 # spends waiting for their file) of selecting a really slow peer
446 # that happens to have a primary share is probably more than
447 # selecting a really fast peer that doesn't have a primary
448 # share. Maybe the servermap could be extended to provide this
449 # information; it could keep track of latency information while
450 # it gathers more important data, and then this routine could
451 # use that to select active readers.
453 # (these and other questions would be easier to answer with a
454 # robust, configurable tahoe-lafs simulator, which modeled node
455 # failures, differences in node speed, and other characteristics
456 # that we expect storage servers to have. You could have
457 # presets for really stable grids (like allmydata.com),
458 # friendnets, make it easy to configure your own settings, and
459 # then simulate the effect of big changes on these use cases
460 # instead of just reasoning about what the effect might be. Out
461 # of scope for MDMF, though.)
463 # XXX: Why don't format= log messages work here?
465 known_shnums = set(self.remaining_sharemap.keys())
466 used_shnums = set([r.shnum for r in self._active_readers])
467 unused_shnums = known_shnums - used_shnums
470 new_shnums = unused_shnums # use them all
471 elif len(self._active_readers) < self._required_shares:
473 more = self._required_shares - len(self._active_readers)
474 # We favor lower numbered shares, since FEC is faster with
475 # primary shares than with other shares, and lower-numbered
476 # shares are more likely to be primary than higher numbered
478 new_shnums = sorted(unused_shnums)[:more]
479 if len(new_shnums) < more:
480 # We don't have enough readers to retrieve the file; fail.
481 self._raise_notenoughshareserror()
485 self.log("adding %d new peers to the active list" % len(new_shnums))
486 for shnum in new_shnums:
487 reader = self.readers[shnum]
488 self._active_readers.append(reader)
489 self.log("added reader for share %d" % shnum)
490 # Each time we add a reader, we check to see if we need the
491 # private key. If we do, we politely ask for it and then continue
492 # computing. If we find that we haven't gotten it at the end of
493 # segment decoding, then we'll take more drastic measures.
494 if self._need_privkey and not self._node.is_readonly():
495 d = reader.get_encprivkey()
496 d.addCallback(self._try_to_validate_privkey, reader)
497 # XXX: don't just drop the Deferred. We need error-reporting
498 # but not flow-control here.
499 assert len(self._active_readers) >= self._required_shares
501 def _try_to_validate_prefix(self, prefix, reader):
503 I check that the prefix returned by a candidate server for
504 retrieval matches the prefix that the servermap knows about
505 (and, hence, the prefix that was validated earlier). If it does,
506 I return True, which means that I approve of the use of the
507 candidate server for segment retrieval. If it doesn't, I return
508 False, which means that another server must be chosen.
518 offsets_tuple) = self.verinfo
519 if known_prefix != prefix:
520 self.log("prefix from share %d doesn't match" % reader.shnum)
521 raise UncoordinatedWriteError("Mismatched prefix -- this could "
522 "indicate an uncoordinated write")
523 # Otherwise, we're okay -- no issues.
526 def _remove_reader(self, reader):
528 At various points, we will wish to remove a peer from
529 consideration and/or use. These include, but are not necessarily
532 - A connection error.
533 - A mismatched prefix (that is, a prefix that does not match
534 our conception of the version information string).
535 - A failing block hash, salt hash, or share hash, which can
536 indicate disk failure/bit flips, or network trouble.
538 This method will do that. I will make sure that the
539 (shnum,reader) combination represented by my reader argument is
540 not used for anything else during this download. I will not
541 advise the reader of any corruption, something that my callers
542 may wish to do on their own.
544 # TODO: When you're done writing this, see if this is ever
545 # actually used for something that _mark_bad_share isn't. I have
546 # a feeling that they will be used for very similar things, and
547 # that having them both here is just going to be an epic amount
548 # of code duplication.
550 # (well, okay, not epic, but meaningful)
551 self.log("removing reader %s" % reader)
552 # Remove the reader from _active_readers
553 self._active_readers.remove(reader)
554 # TODO: self.readers.remove(reader)?
555 for shnum in list(self.remaining_sharemap.keys()):
556 self.remaining_sharemap.discard(shnum, reader.peerid)
559 def _mark_bad_share(self, reader, f):
561 I mark the (peerid, shnum) encapsulated by my reader argument as
562 a bad share, which means that it will not be used anywhere else.
564 There are several reasons to want to mark something as a bad
565 share. These include:
567 - A connection error to the peer.
568 - A mismatched prefix (that is, a prefix that does not match
569 our local conception of the version information string).
570 - A failing block hash, salt hash, share hash, or other
573 This method will ensure that readers that we wish to mark bad
574 (for these reasons or other reasons) are not used for the rest
575 of the download. Additionally, it will attempt to tell the
576 remote peer (with no guarantee of success) that its share is
579 self.log("marking share %d on server %s as bad" % \
580 (reader.shnum, reader))
581 prefix = self.verinfo[-2]
582 self.servermap.mark_bad_share(reader.peerid,
585 self._remove_reader(reader)
586 self._bad_shares.add((reader.peerid, reader.shnum, f))
587 self._status.problems[reader.peerid] = f
588 self._last_failure = f
589 self.notify_server_corruption(reader.peerid, reader.shnum,
593 def _download_current_segment(self):
595 I download, validate, decode, decrypt, and assemble the segment
596 that this Retrieve is currently responsible for downloading.
598 assert len(self._active_readers) >= self._required_shares
599 if self._current_segment > self._last_segment:
600 # No more segments to download, we're done.
601 self.log("got plaintext, done")
603 self.log("on segment %d of %d" %
604 (self._current_segment + 1, self._num_segments))
605 d = self._process_segment(self._current_segment)
606 d.addCallback(lambda ign: self.loop())
609 def _process_segment(self, segnum):
611 I download, validate, decode, and decrypt one segment of the
612 file that this Retrieve is retrieving. This means coordinating
613 the process of getting k blocks of that file, validating them,
614 assembling them into one segment with the decoder, and then
617 self.log("processing segment %d" % segnum)
619 # TODO: The old code uses a marker. Should this code do that
620 # too? What did the Marker do?
621 assert len(self._active_readers) >= self._required_shares
623 # We need to ask each of our active readers for its block and
624 # salt. We will then validate those. If validation is
625 # successful, we will assemble the results into plaintext.
627 for reader in self._active_readers:
628 started = time.time()
629 d = reader.get_block_and_salt(segnum)
630 d2 = self._get_needed_hashes(reader, segnum)
631 dl = defer.DeferredList([d, d2], consumeErrors=True)
632 dl.addCallback(self._validate_block, segnum, reader, started)
633 dl.addErrback(self._validation_or_decoding_failed, [reader])
635 dl = defer.DeferredList(ds)
637 dl.addCallback(lambda ignored: "")
638 dl.addCallback(self._set_segment)
640 dl.addCallback(self._maybe_decode_and_decrypt_segment, segnum)
644 def _maybe_decode_and_decrypt_segment(self, blocks_and_salts, segnum):
646 I take the results of fetching and validating the blocks from a
647 callback chain in another method. If the results are such that
648 they tell me that validation and fetching succeeded without
649 incident, I will proceed with decoding and decryption.
650 Otherwise, I will do nothing.
652 self.log("trying to decode and decrypt segment %d" % segnum)
654 for block_and_salt in blocks_and_salts:
655 if not block_and_salt[0] or block_and_salt[1] == None:
656 self.log("some validation operations failed; not proceeding")
660 self.log("everything looks ok, building segment %d" % segnum)
661 d = self._decode_blocks(blocks_and_salts, segnum)
662 d.addCallback(self._decrypt_segment)
663 d.addErrback(self._validation_or_decoding_failed,
664 self._active_readers)
665 # check to see whether we've been paused before writing
667 d.addCallback(self._check_for_paused)
668 d.addCallback(self._set_segment)
671 return defer.succeed(None)
674 def _set_segment(self, segment):
676 Given a plaintext segment, I register that segment with the
677 target that is handling the file download.
679 self.log("got plaintext for segment %d" % self._current_segment)
680 if self._current_segment == self._start_segment:
681 # We're on the first segment. It's possible that we want
682 # only some part of the end of this segment, and that we
683 # just downloaded the whole thing to get that part. If so,
684 # we need to account for that and give the reader just the
685 # data that they want.
686 n = self._offset % self._segment_size
687 self.log("stripping %d bytes off of the first segment" % n)
688 self.log("original segment length: %d" % len(segment))
689 segment = segment[n:]
690 self.log("new segment length: %d" % len(segment))
692 if self._current_segment == self._last_segment and self._read_length is not None:
693 # We're on the last segment. It's possible that we only want
694 # part of the beginning of this segment, and that we
695 # downloaded the whole thing anyway. Make sure to give the
696 # caller only the portion of the segment that they want to
698 extra = self._read_length
699 if self._start_segment != self._last_segment:
700 extra -= self._segment_size - \
701 (self._offset % self._segment_size)
702 extra %= self._segment_size
703 self.log("original segment length: %d" % len(segment))
704 segment = segment[:extra]
705 self.log("new segment length: %d" % len(segment))
706 self.log("only taking %d bytes of the last segment" % extra)
709 self._consumer.write(segment)
711 # we don't care about the plaintext if we are doing a verify.
713 self._current_segment += 1
716 def _validation_or_decoding_failed(self, f, readers):
718 I am called when a block or a salt fails to correctly validate, or when
719 the decryption or decoding operation fails for some reason. I react to
720 this failure by notifying the remote server of corruption, and then
721 removing the remote peer from further activity.
723 assert isinstance(readers, list)
724 bad_shnums = [reader.shnum for reader in readers]
726 self.log("validation or decoding failed on share(s) %s, peer(s) %s "
727 ", segment %d: %s" % \
728 (bad_shnums, readers, self._current_segment, str(f)))
729 for reader in readers:
730 self._mark_bad_share(reader, f)
734 def _validate_block(self, results, segnum, reader, started):
736 I validate a block from one share on a remote server.
738 # Grab the part of the block hash tree that is necessary to
739 # validate this block, then generate the block hash root.
740 self.log("validating share %d for segment %d" % (reader.shnum,
742 elapsed = time.time() - started
743 self._status.add_fetch_timing(reader.peerid, elapsed)
744 self._set_current_status("validating blocks")
745 # Did we fail to fetch either of the things that we were
746 # supposed to? Fail if so.
747 if not results[0][0] and results[1][0]:
748 # handled by the errback handler.
750 # These all get batched into one query, so the resulting
751 # failure should be the same for all of them, so we can just
753 assert isinstance(results[0][1], failure.Failure)
756 raise CorruptShareError(reader.peerid,
758 "Connection error: %s" % str(f))
760 block_and_salt, block_and_sharehashes = results
761 block, salt = block_and_salt[1]
762 blockhashes, sharehashes = block_and_sharehashes[1]
764 blockhashes = dict(enumerate(blockhashes[1]))
765 self.log("the reader gave me the following blockhashes: %s" % \
767 self.log("the reader gave me the following sharehashes: %s" % \
768 sharehashes[1].keys())
769 bht = self._block_hash_trees[reader.shnum]
771 if bht.needed_hashes(segnum, include_leaf=True):
773 bht.set_hashes(blockhashes)
774 except (hashtree.BadHashError, hashtree.NotEnoughHashesError, \
776 raise CorruptShareError(reader.peerid,
778 "block hash tree failure: %s" % e)
780 if self._version == MDMF_VERSION:
781 blockhash = hashutil.block_hash(salt + block)
783 blockhash = hashutil.block_hash(block)
784 # If this works without an error, then validation is
787 bht.set_hashes(leaves={segnum: blockhash})
788 except (hashtree.BadHashError, hashtree.NotEnoughHashesError, \
790 raise CorruptShareError(reader.peerid,
792 "block hash tree failure: %s" % e)
794 # Reaching this point means that we know that this segment
795 # is correct. Now we need to check to see whether the share
796 # hash chain is also correct.
797 # SDMF wrote share hash chains that didn't contain the
798 # leaves, which would be produced from the block hash tree.
799 # So we need to validate the block hash tree first. If
800 # successful, then bht[0] will contain the root for the
801 # shnum, which will be a leaf in the share hash tree, which
802 # will allow us to validate the rest of the tree.
803 if self.share_hash_tree.needed_hashes(reader.shnum,
804 include_leaf=True) or \
807 self.share_hash_tree.set_hashes(hashes=sharehashes[1],
808 leaves={reader.shnum: bht[0]})
809 except (hashtree.BadHashError, hashtree.NotEnoughHashesError, \
811 raise CorruptShareError(reader.peerid,
813 "corrupt hashes: %s" % e)
815 self.log('share %d is valid for segment %d' % (reader.shnum,
817 return {reader.shnum: (block, salt)}
820 def _get_needed_hashes(self, reader, segnum):
822 I get the hashes needed to validate segnum from the reader, then return
823 to my caller when this is done.
825 bht = self._block_hash_trees[reader.shnum]
826 needed = bht.needed_hashes(segnum, include_leaf=True)
827 # The root of the block hash tree is also a leaf in the share
828 # hash tree. So we don't need to fetch it from the remote
829 # server. In the case of files with one segment, this means that
830 # we won't fetch any block hash tree from the remote server,
831 # since the hash of each share of the file is the entire block
832 # hash tree, and is a leaf in the share hash tree. This is fine,
833 # since any share corruption will be detected in the share hash
836 self.log("getting blockhashes for segment %d, share %d: %s" % \
837 (segnum, reader.shnum, str(needed)))
838 d1 = reader.get_blockhashes(needed, force_remote=True)
839 if self.share_hash_tree.needed_hashes(reader.shnum):
840 need = self.share_hash_tree.needed_hashes(reader.shnum)
841 self.log("also need sharehashes for share %d: %s" % (reader.shnum,
843 d2 = reader.get_sharehashes(need, force_remote=True)
845 d2 = defer.succeed({}) # the logic in the next method
847 dl = defer.DeferredList([d1, d2], consumeErrors=True)
851 def _decode_blocks(self, blocks_and_salts, segnum):
853 I take a list of k blocks and salts, and decode that into a
854 single encrypted segment.
857 # We want to merge our dictionaries to the form
858 # {shnum: blocks_and_salts}
860 # The dictionaries come from validate block that way, so we just
861 # need to merge them.
862 for block_and_salt in blocks_and_salts:
863 d.update(block_and_salt[1])
865 # All of these blocks should have the same salt; in SDMF, it is
866 # the file-wide IV, while in MDMF it is the per-segment salt. In
867 # either case, we just need to get one of them and use it.
869 # d.items()[0] is like (shnum, (block, salt))
870 # d.items()[0][1] is like (block, salt)
871 # d.items()[0][1][1] is the salt.
872 salt = d.items()[0][1][1]
873 # Next, extract just the blocks from the dict. We'll use the
874 # salt in the next step.
875 share_and_shareids = [(k, v[0]) for k, v in d.items()]
876 d2 = dict(share_and_shareids)
879 for shareid, share in d2.items():
880 shareids.append(shareid)
883 self._set_current_status("decoding")
884 started = time.time()
885 assert len(shareids) >= self._required_shares, len(shareids)
886 # zfec really doesn't want extra shares
887 shareids = shareids[:self._required_shares]
888 shares = shares[:self._required_shares]
889 self.log("decoding segment %d" % segnum)
890 if segnum == self._num_segments - 1:
891 d = defer.maybeDeferred(self._tail_decoder.decode, shares, shareids)
893 d = defer.maybeDeferred(self._segment_decoder.decode, shares, shareids)
894 def _process(buffers):
895 segment = "".join(buffers)
896 self.log(format="now decoding segment %(segnum)s of %(numsegs)s",
898 numsegs=self._num_segments,
900 self.log(" joined length %d, datalength %d" %
901 (len(segment), self._data_length))
902 if segnum == self._num_segments - 1:
903 size_to_use = self._tail_data_size
905 size_to_use = self._segment_size
906 segment = segment[:size_to_use]
907 self.log(" segment len=%d" % len(segment))
908 self._status.accumulate_decode_time(time.time() - started)
910 d.addCallback(_process)
914 def _decrypt_segment(self, segment_and_salt):
916 I take a single segment and its salt, and decrypt it. I return
917 the plaintext of the segment that is in my argument.
919 segment, salt = segment_and_salt
920 self._set_current_status("decrypting")
921 self.log("decrypting segment %d" % self._current_segment)
922 started = time.time()
923 key = hashutil.ssk_readkey_data_hash(salt, self._node.get_readkey())
925 plaintext = decryptor.process(segment)
926 self._status.accumulate_decrypt_time(time.time() - started)
930 def notify_server_corruption(self, peerid, shnum, reason):
931 ss = self.servermap.connections[peerid]
932 ss.callRemoteOnly("advise_corrupt_share",
933 "mutable", self._storage_index, shnum, reason)
936 def _try_to_validate_privkey(self, enc_privkey, reader):
937 alleged_privkey_s = self._node._decrypt_privkey(enc_privkey)
938 alleged_writekey = hashutil.ssk_writekey_hash(alleged_privkey_s)
939 if alleged_writekey != self._node.get_writekey():
940 self.log("invalid privkey from %s shnum %d" %
941 (reader, reader.shnum),
942 level=log.WEIRD, umid="YIw4tA")
944 self.servermap.mark_bad_share(reader.peerid, reader.shnum,
946 e = CorruptShareError(reader.peerid,
949 f = failure.Failure(e)
950 self._bad_shares.add((reader.peerid, reader.shnum, f))
954 self.log("got valid privkey from shnum %d on reader %s" %
955 (reader.shnum, reader))
956 privkey = rsa.create_signing_key_from_string(alleged_privkey_s)
957 self._node._populate_encprivkey(enc_privkey)
958 self._node._populate_privkey(privkey)
959 self._need_privkey = False
965 I am called by _download_current_segment when the download process
966 has finished successfully. After making some useful logging
967 statements, I return the decrypted contents to the owner of this
968 Retrieve object through self._done_deferred.
970 self._running = False
971 self._status.set_active(False)
973 self._status.timings['total'] = now - self._started
974 self._status.timings['fetch'] = now - self._started_fetching
975 self._status.set_status("Finished")
976 self._status.set_progress(1.0)
978 # remember the encoding parameters, use them again next time
979 (seqnum, root_hash, IV, segsize, datalength, k, N, prefix,
980 offsets_tuple) = self.verinfo
981 self._node._populate_required_shares(k)
982 self._node._populate_total_shares(N)
985 ret = list(self._bad_shares)
986 self.log("done verifying, found %d bad shares" % len(ret))
988 # TODO: upload status here?
990 self._consumer.unregisterProducer()
991 eventually(self._done_deferred.callback, ret)
994 def _raise_notenoughshareserror(self):
996 I am called by _activate_enough_peers when there are not enough
997 active peers left to complete the download. After making some
998 useful logging statements, I throw an exception to that effect
999 to the caller of this Retrieve object through
1000 self._done_deferred.
1003 format = ("ran out of peers: "
1004 "have %(have)d of %(total)d segments "
1005 "found %(bad)d bad shares "
1006 "encoding %(k)d-of-%(n)d")
1007 args = {"have": self._current_segment,
1008 "total": self._num_segments,
1009 "need": self._last_segment,
1010 "k": self._required_shares,
1011 "n": self._total_shares,
1012 "bad": len(self._bad_shares)}
1013 raise NotEnoughSharesError("%s, last failure: %s" %
1014 (format % args, str(self._last_failure)))
1016 def _error(self, f):
1017 # all errors, including NotEnoughSharesError, land here
1018 self._running = False
1019 self._status.set_active(False)
1021 self._status.timings['total'] = now - self._started
1022 self._status.timings['fetch'] = now - self._started_fetching
1023 self._status.set_status("Failed")
1024 eventually(self._done_deferred.errback, f)