import os, struct, time from itertools import count from zope.interface import implements from twisted.internet import defer from twisted.python import failure from allmydata.interfaces import IPublishStatus, FileTooLargeError from allmydata.util import base32, hashutil, mathutil, idlib, log from allmydata import hashtree, codec, storage from pycryptopp.cipher.aes import AES from foolscap.eventual import eventually from common import MODE_WRITE, DictOfSets, \ UncoordinatedWriteError, NotEnoughServersError from servermap import ServerMap from layout import pack_prefix, pack_share, unpack_header, pack_checkstring, \ unpack_checkstring, SIGNED_PREFIX class PublishStatus: implements(IPublishStatus) statusid_counter = count(0) def __init__(self): self.timings = {} self.timings["send_per_server"] = {} self.servermap = None self.problems = {} self.active = True self.storage_index = None self.helper = False self.encoding = ("?", "?") self.size = None self.status = "Not started" self.progress = 0.0 self.counter = self.statusid_counter.next() self.started = time.time() def add_per_server_time(self, peerid, elapsed): if peerid not in self.timings["send_per_server"]: self.timings["send_per_server"][peerid] = [] self.timings["send_per_server"][peerid].append(elapsed) def get_started(self): return self.started def get_storage_index(self): return self.storage_index def get_encoding(self): return self.encoding def using_helper(self): return self.helper def get_servermap(self): return self.servermap def get_size(self): return self.size def get_status(self): return self.status def get_progress(self): return self.progress def get_active(self): return self.active def get_counter(self): return self.counter def set_storage_index(self, si): self.storage_index = si def set_helper(self, helper): self.helper = helper def set_servermap(self, servermap): self.servermap = servermap def set_encoding(self, k, n): self.encoding = (k, n) def set_size(self, size): self.size = size def set_status(self, status): self.status = status def set_progress(self, value): self.progress = value def set_active(self, value): self.active = value class Publish: """I represent a single act of publishing the mutable file to the grid. I will only publish my data if the servermap I am using still represents the current state of the world. To make the initial publish, set servermap to None. """ # we limit the segment size as usual to constrain our memory footprint. # The max segsize is higher for mutable files, because we want to support # dirnodes with up to 10k children, and each child uses about 330 bytes. # If you actually put that much into a directory you'll be using a # footprint of around 14MB, which is higher than we'd like, but it is # more important right now to support large directories than to make # memory usage small when you use them. Once we implement MDMF (with # multiple segments), we will drop this back down, probably to 128KiB. MAX_SEGMENT_SIZE = 3500000 def __init__(self, filenode, servermap): self._node = filenode self._servermap = servermap self._storage_index = self._node.get_storage_index() self._log_prefix = prefix = storage.si_b2a(self._storage_index)[:5] num = self._node._client.log("Publish(%s): starting" % prefix) self._log_number = num self._running = True self._status = PublishStatus() self._status.set_storage_index(self._storage_index) self._status.set_helper(False) self._status.set_progress(0.0) self._status.set_active(True) def get_status(self): return self._status def log(self, *args, **kwargs): if 'parent' not in kwargs: kwargs['parent'] = self._log_number return log.msg(*args, **kwargs) def publish(self, newdata): """Publish the filenode's current contents. Returns a Deferred that fires (with None) when the publish has done as much work as it's ever going to do, or errbacks with ConsistencyError if it detects a simultaneous write. """ # 1: generate shares (SDMF: files are small, so we can do it in RAM) # 2: perform peer selection, get candidate servers # 2a: send queries to n+epsilon servers, to determine current shares # 2b: based upon responses, create target map # 3: send slot_testv_and_readv_and_writev messages # 4: as responses return, update share-dispatch table # 4a: may need to run recovery algorithm # 5: when enough responses are back, we're done self.log("starting publish, datalen is %s" % len(newdata)) if len(newdata) > self.MAX_SEGMENT_SIZE: raise FileTooLargeError("SDMF is limited to one segment, and " "%d > %d" % (len(newdata), self.MAX_SEGMENT_SIZE)) self._status.set_size(len(newdata)) self._status.set_status("Started") self._started = time.time() self.done_deferred = defer.Deferred() self._writekey = self._node.get_writekey() assert self._writekey, "need write capability to publish" # first, which servers will we publish to? We require that the # servermap was updated in MODE_WRITE, so we can depend upon the # peerlist computed by that process instead of computing our own. if self._servermap: assert self._servermap.last_update_mode == MODE_WRITE # we will push a version that is one larger than anything present # in the grid, according to the servermap. self._new_seqnum = self._servermap.highest_seqnum() + 1 else: # If we don't have a servermap, that's because we're doing the # initial publish self._new_seqnum = 1 self._servermap = ServerMap() self._status.set_servermap(self._servermap) self.log(format="new seqnum will be %(seqnum)d", seqnum=self._new_seqnum, level=log.NOISY) # having an up-to-date servermap (or using a filenode that was just # created for the first time) also guarantees that the following # fields are available self.readkey = self._node.get_readkey() self.required_shares = self._node.get_required_shares() assert self.required_shares is not None self.total_shares = self._node.get_total_shares() assert self.total_shares is not None self._status.set_encoding(self.required_shares, self.total_shares) self._pubkey = self._node.get_pubkey() assert self._pubkey self._privkey = self._node.get_privkey() assert self._privkey self._encprivkey = self._node.get_encprivkey() client = self._node._client full_peerlist = client.get_permuted_peers("storage", self._storage_index) self.full_peerlist = full_peerlist # for use later, immutable self.bad_peers = set() # peerids who have errbacked/refused requests self.newdata = newdata self.salt = os.urandom(16) self.setup_encoding_parameters() # if we experience any surprises (writes which were rejected because # our test vector did not match, or shares which we didn't expect to # see), we set this flag and report an UncoordinatedWriteError at the # end of the publish process. self.surprised = False # as a failsafe, refuse to iterate through self.loop more than a # thousand times. self.looplimit = 1000 # we keep track of three tables. The first is our goal: which share # we want to see on which servers. This is initially populated by the # existing servermap. self.goal = set() # pairs of (peerid, shnum) tuples # the second table is our list of outstanding queries: those which # are in flight and may or may not be delivered, accepted, or # acknowledged. Items are added to this table when the request is # sent, and removed when the response returns (or errbacks). self.outstanding = set() # (peerid, shnum) tuples # the third is a table of successes: share which have actually been # placed. These are populated when responses come back with success. # When self.placed == self.goal, we're done. self.placed = set() # (peerid, shnum) tuples # we also keep a mapping from peerid to RemoteReference. Each time we # pull a connection out of the full peerlist, we add it to this for # use later. self.connections = {} # we use the servermap to populate the initial goal: this way we will # try to update each existing share in place. for (peerid, shnum) in self._servermap.servermap: self.goal.add( (peerid, shnum) ) self.connections[peerid] = self._servermap.connections[peerid] # create the shares. We'll discard these as they are delivered. SMDF: # we're allowed to hold everything in memory. self._status.timings["setup"] = time.time() - self._started d = self._encrypt_and_encode() d.addCallback(self._generate_shares) def _start_pushing(res): self._started_pushing = time.time() return res d.addCallback(_start_pushing) d.addCallback(self.loop) # trigger delivery d.addErrback(self._fatal_error) return self.done_deferred def setup_encoding_parameters(self): segment_size = min(self.MAX_SEGMENT_SIZE, len(self.newdata)) # this must be a multiple of self.required_shares segment_size = mathutil.next_multiple(segment_size, self.required_shares) self.segment_size = segment_size if segment_size: self.num_segments = mathutil.div_ceil(len(self.newdata), segment_size) else: self.num_segments = 0 assert self.num_segments in [0, 1,] # SDMF restrictions def _fatal_error(self, f): self.log("error during loop", failure=f, level=log.SCARY) self._done(f) def _update_status(self): self._status.set_status("Sending Shares: %d placed out of %d, " "%d messages outstanding" % (len(self.placed), len(self.goal), len(self.outstanding))) self._status.set_progress(1.0 * len(self.placed) / len(self.goal)) def loop(self, ignored=None): self.log("entering loop", level=log.NOISY) if not self._running: return self.looplimit -= 1 if self.looplimit <= 0: raise RuntimeError("loop limit exceeded") if self.surprised: # don't send out any new shares, just wait for the outstanding # ones to be retired. self.log("currently surprised, so don't send any new shares", level=log.NOISY) else: self.update_goal() # how far are we from our goal? needed = self.goal - self.placed - self.outstanding self._update_status() if needed: # we need to send out new shares self.log(format="need to send %(needed)d new shares", needed=len(needed), level=log.NOISY) self._send_shares(needed) return if self.outstanding: # queries are still pending, keep waiting self.log(format="%(outstanding)d queries still outstanding", outstanding=len(self.outstanding), level=log.NOISY) return # no queries outstanding, no placements needed: we're done self.log("no queries outstanding, no placements needed: done", level=log.OPERATIONAL) now = time.time() elapsed = now - self._started_pushing self._status.timings["push"] = elapsed return self._done(None) def log_goal(self, goal, message=""): logmsg = [message] for (shnum, peerid) in sorted([(s,p) for (p,s) in goal]): logmsg.append("sh%d to [%s]" % (shnum, idlib.shortnodeid_b2a(peerid))) self.log("current goal: %s" % (", ".join(logmsg)), level=log.NOISY) self.log("we are planning to push new seqnum=#%d" % self._new_seqnum, level=log.NOISY) def update_goal(self): # if log.recording_noisy if True: self.log_goal(self.goal, "before update: ") # first, remove any bad peers from our goal self.goal = set([ (peerid, shnum) for (peerid, shnum) in self.goal if peerid not in self.bad_peers ]) # find the homeless shares: homefull_shares = set([shnum for (peerid, shnum) in self.goal]) homeless_shares = set(range(self.total_shares)) - homefull_shares homeless_shares = sorted(list(homeless_shares)) # place them somewhere. We prefer unused servers at the beginning of # the available peer list. if not homeless_shares: return # if an old share X is on a node, put the new share X there too. # TODO: 1: redistribute shares to achieve one-per-peer, by copying # shares from existing peers to new (less-crowded) ones. The # old shares must still be updated. # TODO: 2: move those shares instead of copying them, to reduce future # update work # this is a bit CPU intensive but easy to analyze. We create a sort # order for each peerid. If the peerid is marked as bad, we don't # even put them in the list. Then we care about the number of shares # which have already been assigned to them. After that we care about # their permutation order. old_assignments = DictOfSets() for (peerid, shnum) in self.goal: old_assignments.add(peerid, shnum) peerlist = [] for i, (peerid, ss) in enumerate(self.full_peerlist): if peerid in self.bad_peers: continue entry = (len(old_assignments.get(peerid, [])), i, peerid, ss) peerlist.append(entry) peerlist.sort() if not peerlist: raise NotEnoughServersError("Ran out of non-bad servers") new_assignments = [] # we then index this peerlist with an integer, because we may have to # wrap. We update the goal as we go. i = 0 for shnum in homeless_shares: (ignored1, ignored2, peerid, ss) = peerlist[i] # TODO: if we are forced to send a share to a server that already # has one, we may have two write requests in flight, and the # servermap (which was computed before either request was sent) # won't reflect the new shares, so the second response will cause # us to be surprised ("unexpected share on peer"), causing the # publish to fail with an UncoordinatedWriteError. This is # troublesome but not really a bit problem. Fix it at some point. self.goal.add( (peerid, shnum) ) self.connections[peerid] = ss i += 1 if i >= len(peerlist): i = 0 if True: self.log_goal(self.goal, "after update: ") def _encrypt_and_encode(self): # this returns a Deferred that fires with a list of (sharedata, # sharenum) tuples. TODO: cache the ciphertext, only produce the # shares that we care about. self.log("_encrypt_and_encode") self._status.set_status("Encrypting") started = time.time() key = hashutil.ssk_readkey_data_hash(self.salt, self.readkey) enc = AES(key) crypttext = enc.process(self.newdata) assert len(crypttext) == len(self.newdata) now = time.time() self._status.timings["encrypt"] = now - started started = now # now apply FEC self._status.set_status("Encoding") fec = codec.CRSEncoder() fec.set_params(self.segment_size, self.required_shares, self.total_shares) piece_size = fec.get_block_size() crypttext_pieces = [None] * self.required_shares for i in range(len(crypttext_pieces)): offset = i * piece_size piece = crypttext[offset:offset+piece_size] piece = piece + "\x00"*(piece_size - len(piece)) # padding crypttext_pieces[i] = piece assert len(piece) == piece_size d = fec.encode(crypttext_pieces) def _done_encoding(res): elapsed = time.time() - started self._status.timings["encode"] = elapsed return res d.addCallback(_done_encoding) return d def _generate_shares(self, shares_and_shareids): # this sets self.shares and self.root_hash self.log("_generate_shares") self._status.set_status("Generating Shares") started = time.time() # we should know these by now privkey = self._privkey encprivkey = self._encprivkey pubkey = self._pubkey (shares, share_ids) = shares_and_shareids assert len(shares) == len(share_ids) assert len(shares) == self.total_shares all_shares = {} block_hash_trees = {} share_hash_leaves = [None] * len(shares) for i in range(len(shares)): share_data = shares[i] shnum = share_ids[i] all_shares[shnum] = share_data # build the block hash tree. SDMF has only one leaf. leaves = [hashutil.block_hash(share_data)] t = hashtree.HashTree(leaves) block_hash_trees[shnum] = block_hash_tree = list(t) share_hash_leaves[shnum] = t[0] for leaf in share_hash_leaves: assert leaf is not None share_hash_tree = hashtree.HashTree(share_hash_leaves) share_hash_chain = {} for shnum in range(self.total_shares): needed_hashes = share_hash_tree.needed_hashes(shnum) share_hash_chain[shnum] = dict( [ (i, share_hash_tree[i]) for i in needed_hashes ] ) root_hash = share_hash_tree[0] assert len(root_hash) == 32 self.log("my new root_hash is %s" % base32.b2a(root_hash)) prefix = pack_prefix(self._new_seqnum, root_hash, self.salt, self.required_shares, self.total_shares, self.segment_size, len(self.newdata)) # now pack the beginning of the share. All shares are the same up # to the signature, then they have divergent share hash chains, # then completely different block hash trees + salt + share data, # then they all share the same encprivkey at the end. The sizes # of everything are the same for all shares. sign_started = time.time() signature = privkey.sign(prefix) self._status.timings["sign"] = time.time() - sign_started verification_key = pubkey.serialize() final_shares = {} for shnum in range(self.total_shares): final_share = pack_share(prefix, verification_key, signature, share_hash_chain[shnum], block_hash_trees[shnum], all_shares[shnum], encprivkey) final_shares[shnum] = final_share elapsed = time.time() - started self._status.timings["pack"] = elapsed self.shares = final_shares self.root_hash = root_hash # we also need to build up the version identifier for what we're # pushing. Extract the offsets from one of our shares. assert final_shares offsets = unpack_header(final_shares.values()[0])[-1] offsets_tuple = tuple( [(key,value) for key,value in offsets.items()] ) verinfo = (self._new_seqnum, root_hash, self.salt, self.segment_size, len(self.newdata), self.required_shares, self.total_shares, prefix, offsets_tuple) self.versioninfo = verinfo def _send_shares(self, needed): self.log("_send_shares") # we're finally ready to send out our shares. If we encounter any # surprises here, it's because somebody else is writing at the same # time. (Note: in the future, when we remove the _query_peers() step # and instead speculate about [or remember] which shares are where, # surprises here are *not* indications of UncoordinatedWriteError, # and we'll need to respond to them more gracefully.) # needed is a set of (peerid, shnum) tuples. The first thing we do is # organize it by peerid. peermap = DictOfSets() for (peerid, shnum) in needed: peermap.add(peerid, shnum) # the next thing is to build up a bunch of test vectors. The # semantics of Publish are that we perform the operation if the world # hasn't changed since the ServerMap was constructed (more or less). # For every share we're trying to place, we create a test vector that # tests to see if the server*share still corresponds to the # map. all_tw_vectors = {} # maps peerid to tw_vectors sm = self._servermap.servermap for key in needed: (peerid, shnum) = key if key in sm: # an old version of that share already exists on the # server, according to our servermap. We will create a # request that attempts to replace it. old_versionid, old_timestamp = sm[key] (old_seqnum, old_root_hash, old_salt, old_segsize, old_datalength, old_k, old_N, old_prefix, old_offsets_tuple) = old_versionid old_checkstring = pack_checkstring(old_seqnum, old_root_hash, old_salt) testv = (0, len(old_checkstring), "eq", old_checkstring) else: # add a testv that requires the share not exist #testv = (0, 1, 'eq', "") # Unfortunately, foolscap-0.2.5 has a bug in the way inbound # constraints are handled. If the same object is referenced # multiple times inside the arguments, foolscap emits a # 'reference' token instead of a distinct copy of the # argument. The bug is that these 'reference' tokens are not # accepted by the inbound constraint code. To work around # this, we need to prevent python from interning the # (constant) tuple, by creating a new copy of this vector # each time. This bug is fixed in later versions of foolscap. testv = tuple([0, 1, 'eq', ""]) testvs = [testv] # the write vector is simply the share writev = [(0, self.shares[shnum])] if peerid not in all_tw_vectors: all_tw_vectors[peerid] = {} # maps shnum to (testvs, writevs, new_length) assert shnum not in all_tw_vectors[peerid] all_tw_vectors[peerid][shnum] = (testvs, writev, None) # we read the checkstring back from each share, however we only use # it to detect whether there was a new share that we didn't know # about. The success or failure of the write will tell us whether # there was a collision or not. If there is a collision, the first # thing we'll do is update the servermap, which will find out what # happened. We could conceivably reduce a roundtrip by using the # readv checkstring to populate the servermap, but really we'd have # to read enough data to validate the signatures too, so it wouldn't # be an overall win. read_vector = [(0, struct.calcsize(SIGNED_PREFIX))] # ok, send the messages! self.log("sending %d shares" % len(all_tw_vectors), level=log.NOISY) started = time.time() for (peerid, tw_vectors) in all_tw_vectors.items(): write_enabler = self._node.get_write_enabler(peerid) renew_secret = self._node.get_renewal_secret(peerid) cancel_secret = self._node.get_cancel_secret(peerid) secrets = (write_enabler, renew_secret, cancel_secret) shnums = tw_vectors.keys() for shnum in shnums: self.outstanding.add( (peerid, shnum) ) d = self._do_testreadwrite(peerid, secrets, tw_vectors, read_vector) d.addCallbacks(self._got_write_answer, self._got_write_error, callbackArgs=(peerid, shnums, started), errbackArgs=(peerid, shnums, started)) d.addCallback(self.loop) d.addErrback(self._fatal_error) self._update_status() self.log("%d shares sent" % len(all_tw_vectors), level=log.NOISY) def _do_testreadwrite(self, peerid, secrets, tw_vectors, read_vector): storage_index = self._storage_index ss = self.connections[peerid] #print "SS[%s] is %s" % (idlib.shortnodeid_b2a(peerid), ss), ss.tracker.interfaceName d = ss.callRemote("slot_testv_and_readv_and_writev", storage_index, secrets, tw_vectors, read_vector) return d def _got_write_answer(self, answer, peerid, shnums, started): lp = self.log("_got_write_answer from %s" % idlib.shortnodeid_b2a(peerid)) for shnum in shnums: self.outstanding.discard( (peerid, shnum) ) now = time.time() elapsed = now - started self._status.add_per_server_time(peerid, elapsed) wrote, read_data = answer surprise_shares = set(read_data.keys()) - set(shnums) if surprise_shares: self.log("they had shares %s that we didn't know about" % (list(surprise_shares),), parent=lp, level=log.WEIRD) self.surprised = True if not wrote: # TODO: there are two possibilities. The first is that the server # is full (or just doesn't want to give us any room), which means # we shouldn't ask them again, but is *not* an indication of an # uncoordinated write. The second is that our testv failed, which # *does* indicate an uncoordinated write. We currently don't have # a way to tell these two apart (in fact, the storage server code # doesn't have the option of refusing our share). # # If the server is full, mark the peer as bad (so we don't ask # them again), but don't set self.surprised. The loop() will find # a new server. # # If the testv failed, log it, set self.surprised, but don't # bother adding to self.bad_peers . self.log("our testv failed, so the write did not happen", parent=lp, level=log.WEIRD) self.surprised = True self.bad_peers.add(peerid) # don't ask them again # use the checkstring to add information to the log message for (shnum,readv) in read_data.items(): checkstring = readv[0] (other_seqnum, other_roothash, other_salt) = unpack_checkstring(checkstring) expected_version = self._servermap.version_on_peer(peerid, shnum) if expected_version: (seqnum, root_hash, IV, segsize, datalength, k, N, prefix, offsets_tuple) = expected_version self.log("somebody modified the share on us:" " shnum=%d: I thought they had #%d:R=%s," " but testv reported #%d:R=%s" % (shnum, seqnum, base32.b2a(root_hash)[:4], other_seqnum, base32.b2a(other_roothash)[:4]), parent=lp, level=log.NOISY) # if expected_version==None, then we didn't expect to see a # share on that peer, and the 'surprise_shares' clause above # will have logged it. # self.loop() will take care of finding new homes return for shnum in shnums: self.placed.add( (peerid, shnum) ) # and update the servermap self._servermap.add_new_share(peerid, shnum, self.versioninfo, started) # self.loop() will take care of checking to see if we're done return def _got_write_error(self, f, peerid, shnums, started): for shnum in shnums: self.outstanding.discard( (peerid, shnum) ) self.bad_peers.add(peerid) self.log(format="error while writing shares %(shnums)s to peerid %(peerid)s", shnums=list(shnums), peerid=idlib.shortnodeid_b2a(peerid), failure=f, level=log.UNUSUAL) # self.loop() will take care of checking to see if we're done return def _done(self, res): if not self._running: return self._running = False now = time.time() self._status.timings["total"] = now - self._started self._status.set_active(False) if isinstance(res, failure.Failure): self.log("Publish done, with failure", failure=res, level=log.WEIRD) self._status.set_status("Failed") elif self.surprised: self.log("Publish done, UncoordinatedWriteError", level=log.UNUSUAL) self._status.set_status("UncoordinatedWriteError") # deliver a failure res = failure.Failure(UncoordinatedWriteError()) # TODO: recovery else: self.log("Publish done, success") self._status.set_status("Done") self._status.set_progress(1.0) eventually(self.done_deferred.callback, res)