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, DeadReferenceError, \
10 from allmydata.interfaces import IRetrieveStatus, NotEnoughSharesError, \
11 DownloadStopped, MDMF_VERSION, SDMF_VERSION
12 from allmydata.util import hashutil, log, mathutil, deferredutil
13 from allmydata.util.dictutil import DictOfSets
14 from allmydata import hashtree, codec
15 from allmydata.storage.server import si_b2a
16 from pycryptopp.cipher.aes import AES
17 from pycryptopp.publickey import rsa
19 from allmydata.mutable.common import CorruptShareError, BadShareError, \
20 UncoordinatedWriteError
21 from allmydata.mutable.layout import MDMFSlotReadProxy
24 implements(IRetrieveStatus)
25 statusid_counter = count(0)
28 self.timings["fetch_per_server"] = {}
29 self.timings["decode"] = 0.0
30 self.timings["decrypt"] = 0.0
31 self.timings["cumulative_verify"] = 0.0
34 self.storage_index = None
36 self.encoding = ("?","?")
38 self.status = "Not started"
40 self.counter = self.statusid_counter.next()
41 self.started = time.time()
43 def get_started(self):
45 def get_storage_index(self):
46 return self.storage_index
47 def get_encoding(self):
49 def using_helper(self):
55 def get_progress(self):
59 def get_counter(self):
61 def get_problems(self):
64 def add_fetch_timing(self, server, elapsed):
65 serverid = server.get_serverid()
66 if serverid not in self.timings["fetch_per_server"]:
67 self.timings["fetch_per_server"][serverid] = []
68 self.timings["fetch_per_server"][serverid].append(elapsed)
69 def accumulate_decode_time(self, elapsed):
70 self.timings["decode"] += elapsed
71 def accumulate_decrypt_time(self, elapsed):
72 self.timings["decrypt"] += elapsed
73 def set_storage_index(self, si):
74 self.storage_index = si
75 def set_helper(self, helper):
77 def set_encoding(self, k, n):
78 self.encoding = (k, n)
79 def set_size(self, size):
81 def set_status(self, status):
83 def set_progress(self, value):
85 def set_active(self, value):
87 def add_problem(self, server, f):
88 serverid = server.get_serverid()
89 self._problems[serverid] = f
95 # this class is currently single-use. Eventually (in MDMF) we will make
96 # it multi-use, in which case you can call download(range) multiple
97 # times, and each will have a separate response chain. However the
98 # Retrieve object will remain tied to a specific version of the file, and
99 # will use a single ServerMap instance.
100 implements(IPushProducer)
102 def __init__(self, filenode, storage_broker, servermap, verinfo,
103 fetch_privkey=False, verify=False):
104 self._node = filenode
105 assert self._node.get_pubkey()
106 self._storage_broker = storage_broker
107 self._storage_index = filenode.get_storage_index()
108 assert self._node.get_readkey()
109 self._last_failure = None
110 prefix = si_b2a(self._storage_index)[:5]
111 self._log_number = log.msg("Retrieve(%s): starting" % prefix)
113 self._decoding = False
114 self._bad_shares = set()
116 self.servermap = servermap
117 assert self._node.get_pubkey()
118 self.verinfo = verinfo
119 # during repair, we may be called upon to grab the private key, since
120 # it wasn't picked up during a verify=False checker run, and we'll
121 # need it for repair to generate a new version.
122 self._need_privkey = verify or (fetch_privkey
123 and not self._node.get_privkey())
125 if self._need_privkey:
126 # TODO: Evaluate the need for this. We'll use it if we want
127 # to limit how many queries are on the wire for the privkey
129 self._privkey_query_markers = [] # one Marker for each time we've
130 # tried to get the privkey.
132 # verify means that we are using the downloader logic to verify all
133 # of our shares. This tells the downloader a few things.
135 # 1. We need to download all of the shares.
136 # 2. We don't need to decode or decrypt the shares, since our
137 # caller doesn't care about the plaintext, only the
138 # information about which shares are or are not valid.
139 # 3. When we are validating readers, we need to validate the
140 # signature on the prefix. Do we? We already do this in the
142 self._verify = verify
144 self._status = RetrieveStatus()
145 self._status.set_storage_index(self._storage_index)
146 self._status.set_helper(False)
147 self._status.set_progress(0.0)
148 self._status.set_active(True)
149 (seqnum, root_hash, IV, segsize, datalength, k, N, prefix,
150 offsets_tuple) = self.verinfo
151 self._status.set_size(datalength)
152 self._status.set_encoding(k, N)
154 self._stopped = False
155 self._pause_deferred = None
157 self._read_length = None
158 self.log("got seqnum %d" % self.verinfo[0])
161 def get_status(self):
164 def log(self, *args, **kwargs):
165 if "parent" not in kwargs:
166 kwargs["parent"] = self._log_number
167 if "facility" not in kwargs:
168 kwargs["facility"] = "tahoe.mutable.retrieve"
169 return log.msg(*args, **kwargs)
171 def _set_current_status(self, state):
172 seg = "%d/%d" % (self._current_segment, self._last_segment)
173 self._status.set_status("segment %s (%s)" % (seg, state))
178 def pauseProducing(self):
180 I am called by my download target if we have produced too much
181 data for it to handle. I make the downloader stop producing new
182 data until my resumeProducing method is called.
184 if self._pause_deferred is not None:
187 # fired when the download is unpaused.
188 self._old_status = self._status.get_status()
189 self._set_current_status("paused")
191 self._pause_deferred = defer.Deferred()
194 def resumeProducing(self):
196 I am called by my download target once it is ready to begin
197 receiving data again.
199 if self._pause_deferred is None:
202 p = self._pause_deferred
203 self._pause_deferred = None
204 self._status.set_status(self._old_status)
206 eventually(p.callback, None)
208 def stopProducing(self):
210 self.resumeProducing()
213 def _check_for_paused(self, res):
215 I am called just before a write to the consumer. I return a
216 Deferred that eventually fires with the data that is to be
217 written to the consumer. If the download has not been paused,
218 the Deferred fires immediately. Otherwise, the Deferred fires
219 when the downloader is unpaused.
221 if self._pause_deferred is not None:
223 self._pause_deferred.addCallback(lambda ignored: d.callback(res))
227 def _check_for_stopped(self, res):
229 raise DownloadStopped("our Consumer called stopProducing()")
233 def download(self, consumer=None, offset=0, size=None):
234 assert IConsumer.providedBy(consumer) or self._verify
237 self._consumer = consumer
238 # we provide IPushProducer, so streaming=True, per
240 self._consumer.registerProducer(self, streaming=True)
242 self._done_deferred = defer.Deferred()
243 self._offset = offset
244 self._read_length = size
245 self._setup_download()
246 self._setup_encoding_parameters()
247 self.log("starting download")
248 self._started_fetching = time.time()
249 # The download process beyond this is a state machine.
250 # _add_active_servers will select the servers that we want to use
251 # for the download, and then attempt to start downloading. After
252 # each segment, it will check for doneness, reacting to broken
253 # servers and corrupt shares as necessary. If it runs out of good
254 # servers before downloading all of the segments, _done_deferred
255 # will errback. Otherwise, it will eventually callback with the
256 # contents of the mutable file.
258 return self._done_deferred
261 d = fireEventually(None) # avoid #237 recursion limit problem
262 d.addCallback(lambda ign: self._activate_enough_servers())
263 d.addCallback(lambda ign: self._download_current_segment())
264 # when we're done, _download_current_segment will call _done. If we
265 # aren't, it will call loop() again.
266 d.addErrback(self._error)
268 def _setup_download(self):
269 self._started = time.time()
270 self._status.set_status("Retrieving Shares")
272 # how many shares do we need?
281 offsets_tuple) = self.verinfo
283 # first, which servers can we use?
284 versionmap = self.servermap.make_versionmap()
285 shares = versionmap[self.verinfo]
286 # this sharemap is consumed as we decide to send requests
287 self.remaining_sharemap = DictOfSets()
288 for (shnum, server, timestamp) in shares:
289 self.remaining_sharemap.add(shnum, server)
290 # If the servermap update fetched anything, it fetched at least 1
291 # KiB, so we ask for that much.
292 # TODO: Change the cache methods to allow us to fetch all of the
293 # data that they have, then change this method to do that.
294 any_cache = self._node._read_from_cache(self.verinfo, shnum,
296 reader = MDMFSlotReadProxy(server.get_rref(),
300 reader.server = server
301 self.readers[shnum] = reader
302 assert len(self.remaining_sharemap) >= k
304 self.shares = {} # maps shnum to validated blocks
305 self._active_readers = [] # list of active readers for this dl.
306 self._block_hash_trees = {} # shnum => hashtree
308 # We need one share hash tree for the entire file; its leaves
309 # are the roots of the block hash trees for the shares that
310 # comprise it, and its root is in the verinfo.
311 self.share_hash_tree = hashtree.IncompleteHashTree(N)
312 self.share_hash_tree.set_hashes({0: root_hash})
314 def decode(self, blocks_and_salts, segnum):
316 I am a helper method that the mutable file update process uses
317 as a shortcut to decode and decrypt the segments that it needs
318 to fetch in order to perform a file update. I take in a
319 collection of blocks and salts, and pick some of those to make a
320 segment with. I return the plaintext associated with that
323 # shnum => block hash tree. Unused, but setup_encoding_parameters will
325 self._block_hash_trees = None
326 self._setup_encoding_parameters()
328 # _decode_blocks() expects the output of a gatherResults that
329 # contains the outputs of _validate_block() (each of which is a dict
330 # mapping shnum to (block,salt) bytestrings).
331 d = self._decode_blocks([blocks_and_salts], segnum)
332 d.addCallback(self._decrypt_segment)
336 def _setup_encoding_parameters(self):
338 I set up the encoding parameters, including k, n, the number
339 of segments associated with this file, and the segment decoders.
349 offsets_tuple) = self.verinfo
350 self._required_shares = k
351 self._total_shares = n
352 self._segment_size = segsize
353 self._data_length = datalength
356 self._version = MDMF_VERSION
358 self._version = SDMF_VERSION
360 if datalength and segsize:
361 self._num_segments = mathutil.div_ceil(datalength, segsize)
362 self._tail_data_size = datalength % segsize
364 self._num_segments = 0
365 self._tail_data_size = 0
367 self._segment_decoder = codec.CRSDecoder()
368 self._segment_decoder.set_params(segsize, k, n)
370 if not self._tail_data_size:
371 self._tail_data_size = segsize
373 self._tail_segment_size = mathutil.next_multiple(self._tail_data_size,
374 self._required_shares)
375 if self._tail_segment_size == self._segment_size:
376 self._tail_decoder = self._segment_decoder
378 self._tail_decoder = codec.CRSDecoder()
379 self._tail_decoder.set_params(self._tail_segment_size,
380 self._required_shares,
383 self.log("got encoding parameters: "
386 "%d segments of %d bytes each (%d byte tail segment)" % \
387 (k, n, self._num_segments, self._segment_size,
388 self._tail_segment_size))
390 if self._block_hash_trees is not None:
391 for i in xrange(self._total_shares):
392 # So we don't have to do this later.
393 self._block_hash_trees[i] = hashtree.IncompleteHashTree(self._num_segments)
395 # Our last task is to tell the downloader where to start and
396 # where to stop. We use three parameters for that:
397 # - self._start_segment: the segment that we need to start
399 # - self._current_segment: the next segment that we need to
401 # - self._last_segment: The last segment that we were asked to
404 # We say that the download is complete when
405 # self._current_segment > self._last_segment. We use
406 # self._start_segment and self._last_segment to know when to
407 # strip things off of segments, and how much to strip.
409 self.log("got offset: %d" % self._offset)
410 # our start segment is the first segment containing the
411 # offset we were given.
412 start = self._offset // self._segment_size
414 assert start < self._num_segments
415 self._start_segment = start
416 self.log("got start segment: %d" % self._start_segment)
418 self._start_segment = 0
421 # If self._read_length is None, then we want to read the whole
422 # file. Otherwise, we want to read only part of the file, and
423 # need to figure out where to stop reading.
424 if self._read_length is not None:
425 # our end segment is the last segment containing part of the
426 # segment that we were asked to read.
427 self.log("got read length %d" % self._read_length)
428 if self._read_length != 0:
429 end_data = self._offset + self._read_length
431 # We don't actually need to read the byte at end_data,
432 # but the one before it.
433 end = (end_data - 1) // self._segment_size
435 assert end < self._num_segments
436 self._last_segment = end
438 self._last_segment = self._start_segment
439 self.log("got end segment: %d" % self._last_segment)
441 self._last_segment = self._num_segments - 1
443 self._current_segment = self._start_segment
445 def _activate_enough_servers(self):
447 I populate self._active_readers with enough active readers to
448 retrieve the contents of this mutable file. I am called before
449 downloading starts, and (eventually) after each validation
450 error, connection error, or other problem in the download.
452 # TODO: It would be cool to investigate other heuristics for
453 # reader selection. For instance, the cost (in time the user
454 # spends waiting for their file) of selecting a really slow server
455 # that happens to have a primary share is probably more than
456 # selecting a really fast server that doesn't have a primary
457 # share. Maybe the servermap could be extended to provide this
458 # information; it could keep track of latency information while
459 # it gathers more important data, and then this routine could
460 # use that to select active readers.
462 # (these and other questions would be easier to answer with a
463 # robust, configurable tahoe-lafs simulator, which modeled node
464 # failures, differences in node speed, and other characteristics
465 # that we expect storage servers to have. You could have
466 # presets for really stable grids (like allmydata.com),
467 # friendnets, make it easy to configure your own settings, and
468 # then simulate the effect of big changes on these use cases
469 # instead of just reasoning about what the effect might be. Out
470 # of scope for MDMF, though.)
472 # XXX: Why don't format= log messages work here?
474 known_shnums = set(self.remaining_sharemap.keys())
475 used_shnums = set([r.shnum for r in self._active_readers])
476 unused_shnums = known_shnums - used_shnums
479 new_shnums = unused_shnums # use them all
480 elif len(self._active_readers) < self._required_shares:
482 more = self._required_shares - len(self._active_readers)
483 # We favor lower numbered shares, since FEC is faster with
484 # primary shares than with other shares, and lower-numbered
485 # shares are more likely to be primary than higher numbered
487 new_shnums = sorted(unused_shnums)[:more]
488 if len(new_shnums) < more:
489 # We don't have enough readers to retrieve the file; fail.
490 self._raise_notenoughshareserror()
494 self.log("adding %d new servers to the active list" % len(new_shnums))
495 for shnum in new_shnums:
496 reader = self.readers[shnum]
497 self._active_readers.append(reader)
498 self.log("added reader for share %d" % shnum)
499 # Each time we add a reader, we check to see if we need the
500 # private key. If we do, we politely ask for it and then continue
501 # computing. If we find that we haven't gotten it at the end of
502 # segment decoding, then we'll take more drastic measures.
503 if self._need_privkey and not self._node.is_readonly():
504 d = reader.get_encprivkey()
505 d.addCallback(self._try_to_validate_privkey, reader, reader.server)
506 # XXX: don't just drop the Deferred. We need error-reporting
507 # but not flow-control here.
509 def _try_to_validate_prefix(self, prefix, reader):
511 I check that the prefix returned by a candidate server for
512 retrieval matches the prefix that the servermap knows about
513 (and, hence, the prefix that was validated earlier). If it does,
514 I return True, which means that I approve of the use of the
515 candidate server for segment retrieval. If it doesn't, I return
516 False, which means that another server must be chosen.
526 offsets_tuple) = self.verinfo
527 if known_prefix != prefix:
528 self.log("prefix from share %d doesn't match" % reader.shnum)
529 raise UncoordinatedWriteError("Mismatched prefix -- this could "
530 "indicate an uncoordinated write")
531 # Otherwise, we're okay -- no issues.
534 def _remove_reader(self, reader):
536 At various points, we will wish to remove a server from
537 consideration and/or use. These include, but are not necessarily
540 - A connection error.
541 - A mismatched prefix (that is, a prefix that does not match
542 our conception of the version information string).
543 - A failing block hash, salt hash, or share hash, which can
544 indicate disk failure/bit flips, or network trouble.
546 This method will do that. I will make sure that the
547 (shnum,reader) combination represented by my reader argument is
548 not used for anything else during this download. I will not
549 advise the reader of any corruption, something that my callers
550 may wish to do on their own.
552 # TODO: When you're done writing this, see if this is ever
553 # actually used for something that _mark_bad_share isn't. I have
554 # a feeling that they will be used for very similar things, and
555 # that having them both here is just going to be an epic amount
556 # of code duplication.
558 # (well, okay, not epic, but meaningful)
559 self.log("removing reader %s" % reader)
560 # Remove the reader from _active_readers
561 self._active_readers.remove(reader)
562 # TODO: self.readers.remove(reader)?
563 for shnum in list(self.remaining_sharemap.keys()):
564 self.remaining_sharemap.discard(shnum, reader.server)
567 def _mark_bad_share(self, server, shnum, reader, f):
569 I mark the given (server, shnum) as a bad share, which means that it
570 will not be used anywhere else.
572 There are several reasons to want to mark something as a bad
573 share. These include:
575 - A connection error to the server.
576 - A mismatched prefix (that is, a prefix that does not match
577 our local conception of the version information string).
578 - A failing block hash, salt hash, share hash, or other
581 This method will ensure that readers that we wish to mark bad
582 (for these reasons or other reasons) are not used for the rest
583 of the download. Additionally, it will attempt to tell the
584 remote server (with no guarantee of success) that its share is
587 self.log("marking share %d on server %s as bad" % \
588 (shnum, server.get_name()))
589 prefix = self.verinfo[-2]
590 self.servermap.mark_bad_share(server, shnum, prefix)
591 self._remove_reader(reader)
592 self._bad_shares.add((server, shnum, f))
593 self._status.add_problem(server, f)
594 self._last_failure = f
595 if f.check(BadShareError):
596 self.notify_server_corruption(server, shnum, str(f.value))
599 def _download_current_segment(self):
601 I download, validate, decode, decrypt, and assemble the segment
602 that this Retrieve is currently responsible for downloading.
604 if self._current_segment > self._last_segment:
605 # No more segments to download, we're done.
606 self.log("got plaintext, done")
608 elif self._verify and len(self._active_readers) == 0:
609 self.log("no more good shares, no need to keep verifying")
611 self.log("on segment %d of %d" %
612 (self._current_segment + 1, self._num_segments))
613 d = self._process_segment(self._current_segment)
614 d.addCallback(lambda ign: self.loop())
617 def _process_segment(self, segnum):
619 I download, validate, decode, and decrypt one segment of the
620 file that this Retrieve is retrieving. This means coordinating
621 the process of getting k blocks of that file, validating them,
622 assembling them into one segment with the decoder, and then
625 self.log("processing segment %d" % segnum)
627 # TODO: The old code uses a marker. Should this code do that
628 # too? What did the Marker do?
630 # We need to ask each of our active readers for its block and
631 # salt. We will then validate those. If validation is
632 # successful, we will assemble the results into plaintext.
634 for reader in self._active_readers:
635 started = time.time()
636 d1 = reader.get_block_and_salt(segnum)
637 d2,d3 = self._get_needed_hashes(reader, segnum)
638 d = deferredutil.gatherResults([d1,d2,d3])
639 d.addCallback(self._validate_block, segnum, reader, reader.server, started)
640 # _handle_bad_share takes care of recoverable errors (by dropping
641 # that share and returning None). Any other errors (i.e. code
642 # bugs) are passed through and cause the retrieve to fail.
643 d.addErrback(self._handle_bad_share, [reader])
645 dl = deferredutil.gatherResults(ds)
647 dl.addCallback(lambda ignored: "")
648 dl.addCallback(self._set_segment)
650 dl.addCallback(self._maybe_decode_and_decrypt_segment, segnum)
654 def _maybe_decode_and_decrypt_segment(self, results, segnum):
656 I take the results of fetching and validating the blocks from
657 _process_segment. If validation and fetching succeeded without
658 incident, I will proceed with decoding and decryption. Otherwise, I
661 self.log("trying to decode and decrypt segment %d" % segnum)
663 # 'results' is the output of a gatherResults set up in
664 # _process_segment(). Each component Deferred will either contain the
665 # non-Failure output of _validate_block() for a single block (i.e.
666 # {segnum:(block,salt)}), or None if _validate_block threw an
667 # exception and _validation_or_decoding_failed handled it (by
668 # dropping that server).
671 self.log("some validation operations failed; not proceeding")
672 return defer.succeed(None)
673 self.log("everything looks ok, building segment %d" % segnum)
674 d = self._decode_blocks(results, segnum)
675 d.addCallback(self._decrypt_segment)
676 # check to see whether we've been paused before writing
678 d.addCallback(self._check_for_paused)
679 d.addCallback(self._check_for_stopped)
680 d.addCallback(self._set_segment)
684 def _set_segment(self, segment):
686 Given a plaintext segment, I register that segment with the
687 target that is handling the file download.
689 self.log("got plaintext for segment %d" % self._current_segment)
690 if self._current_segment == self._start_segment:
691 # We're on the first segment. It's possible that we want
692 # only some part of the end of this segment, and that we
693 # just downloaded the whole thing to get that part. If so,
694 # we need to account for that and give the reader just the
695 # data that they want.
696 n = self._offset % self._segment_size
697 self.log("stripping %d bytes off of the first segment" % n)
698 self.log("original segment length: %d" % len(segment))
699 segment = segment[n:]
700 self.log("new segment length: %d" % len(segment))
702 if self._current_segment == self._last_segment and self._read_length is not None:
703 # We're on the last segment. It's possible that we only want
704 # part of the beginning of this segment, and that we
705 # downloaded the whole thing anyway. Make sure to give the
706 # caller only the portion of the segment that they want to
708 extra = self._read_length
709 if self._start_segment != self._last_segment:
710 extra -= self._segment_size - \
711 (self._offset % self._segment_size)
712 extra %= self._segment_size
713 self.log("original segment length: %d" % len(segment))
714 segment = segment[:extra]
715 self.log("new segment length: %d" % len(segment))
716 self.log("only taking %d bytes of the last segment" % extra)
719 self._consumer.write(segment)
721 # we don't care about the plaintext if we are doing a verify.
723 self._current_segment += 1
726 def _handle_bad_share(self, f, readers):
728 I am called when a block or a salt fails to correctly validate, or when
729 the decryption or decoding operation fails for some reason. I react to
730 this failure by notifying the remote server of corruption, and then
731 removing the remote server from further activity.
733 # these are the errors we can tolerate: by giving up on this share
734 # and finding others to replace it. Any other errors (i.e. coding
735 # bugs) are re-raised, causing the download to fail.
736 f.trap(DeadReferenceError, RemoteException, BadShareError)
738 # DeadReferenceError happens when we try to fetch data from a server
739 # that has gone away. RemoteException happens if the server had an
740 # internal error. BadShareError encompasses: (UnknownVersionError,
741 # LayoutInvalid, struct.error) which happen when we get obviously
742 # wrong data, and CorruptShareError which happens later, when we
743 # perform integrity checks on the data.
745 assert isinstance(readers, list)
746 bad_shnums = [reader.shnum for reader in readers]
748 self.log("validation or decoding failed on share(s) %s, server(s) %s "
749 ", segment %d: %s" % \
750 (bad_shnums, readers, self._current_segment, str(f)))
751 for reader in readers:
752 self._mark_bad_share(reader.server, reader.shnum, reader, f)
756 def _validate_block(self, results, segnum, reader, server, started):
758 I validate a block from one share on a remote server.
760 # Grab the part of the block hash tree that is necessary to
761 # validate this block, then generate the block hash root.
762 self.log("validating share %d for segment %d" % (reader.shnum,
764 elapsed = time.time() - started
765 self._status.add_fetch_timing(server, elapsed)
766 self._set_current_status("validating blocks")
768 block_and_salt, blockhashes, sharehashes = results
769 block, salt = block_and_salt
771 blockhashes = dict(enumerate(blockhashes))
772 self.log("the reader gave me the following blockhashes: %s" % \
774 self.log("the reader gave me the following sharehashes: %s" % \
776 bht = self._block_hash_trees[reader.shnum]
778 if bht.needed_hashes(segnum, include_leaf=True):
780 bht.set_hashes(blockhashes)
781 except (hashtree.BadHashError, hashtree.NotEnoughHashesError, \
783 raise CorruptShareError(server,
785 "block hash tree failure: %s" % e)
787 if self._version == MDMF_VERSION:
788 blockhash = hashutil.block_hash(salt + block)
790 blockhash = hashutil.block_hash(block)
791 # If this works without an error, then validation is
794 bht.set_hashes(leaves={segnum: blockhash})
795 except (hashtree.BadHashError, hashtree.NotEnoughHashesError, \
797 raise CorruptShareError(server,
799 "block hash tree failure: %s" % e)
801 # Reaching this point means that we know that this segment
802 # is correct. Now we need to check to see whether the share
803 # hash chain is also correct.
804 # SDMF wrote share hash chains that didn't contain the
805 # leaves, which would be produced from the block hash tree.
806 # So we need to validate the block hash tree first. If
807 # successful, then bht[0] will contain the root for the
808 # shnum, which will be a leaf in the share hash tree, which
809 # will allow us to validate the rest of the tree.
811 self.share_hash_tree.set_hashes(hashes=sharehashes,
812 leaves={reader.shnum: bht[0]})
813 except (hashtree.BadHashError, hashtree.NotEnoughHashesError, \
815 raise CorruptShareError(server,
817 "corrupt hashes: %s" % e)
819 self.log('share %d is valid for segment %d' % (reader.shnum,
821 return {reader.shnum: (block, salt)}
824 def _get_needed_hashes(self, reader, segnum):
826 I get the hashes needed to validate segnum from the reader, then return
827 to my caller when this is done.
829 bht = self._block_hash_trees[reader.shnum]
830 needed = bht.needed_hashes(segnum, include_leaf=True)
831 # The root of the block hash tree is also a leaf in the share
832 # hash tree. So we don't need to fetch it from the remote
833 # server. In the case of files with one segment, this means that
834 # we won't fetch any block hash tree from the remote server,
835 # since the hash of each share of the file is the entire block
836 # hash tree, and is a leaf in the share hash tree. This is fine,
837 # since any share corruption will be detected in the share hash
840 self.log("getting blockhashes for segment %d, share %d: %s" % \
841 (segnum, reader.shnum, str(needed)))
842 d1 = reader.get_blockhashes(needed, force_remote=True)
843 if self.share_hash_tree.needed_hashes(reader.shnum):
844 need = self.share_hash_tree.needed_hashes(reader.shnum)
845 self.log("also need sharehashes for share %d: %s" % (reader.shnum,
847 d2 = reader.get_sharehashes(need, force_remote=True)
849 d2 = defer.succeed({}) # the logic in the next method
854 def _decode_blocks(self, results, segnum):
856 I take a list of k blocks and salts, and decode that into a
857 single encrypted segment.
859 # 'results' is one or more dicts (each {shnum:(block,salt)}), and we
860 # want to merge them all
861 blocks_and_salts = {}
863 blocks_and_salts.update(d)
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 = blocks_and_salts.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 blocks_and_salts.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, server, shnum, reason):
931 rref = server.get_rref()
932 rref.callRemoteOnly("advise_corrupt_share",
933 "mutable", self._storage_index, shnum, reason)
936 def _try_to_validate_privkey(self, enc_privkey, reader, server):
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(server, reader.shnum,
946 e = CorruptShareError(server,
949 f = failure.Failure(e)
950 self._bad_shares.add((server, 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 = 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_servers when there are not enough
997 active servers 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 servers: "
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)