2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
12 Hash = StringConstraint(maxLength=HASH_SIZE,
13 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
14 Nodeid = StringConstraint(maxLength=20,
15 minLength=20) # binary format 20-byte SHA1 hash
16 FURL = StringConstraint(1000)
17 StorageIndex = StringConstraint(16)
18 URI = StringConstraint(300) # kind of arbitrary
20 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
22 DEFAULT_MAX_SEGMENT_SIZE = 128*1024
24 ShareData = StringConstraint(None)
25 URIExtensionData = StringConstraint(1000)
26 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
28 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
29 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
30 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
31 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
33 class RIBucketWriter(RemoteInterface):
34 """ Objects of this kind live on the server side. """
35 def write(offset=Offset, data=ShareData):
40 If the data that has been written is incomplete or inconsistent then
41 the server will throw the data away, else it will store it for future
47 """Abandon all the data that has been written.
51 class RIBucketReader(RemoteInterface):
52 def read(offset=Offset, length=ReadSize):
55 def advise_corrupt_share(reason=str):
56 """Clients who discover hash failures in shares that they have
57 downloaded from me will use this method to inform me about the
58 failures. I will record their concern so that my operator can
59 manually inspect the shares in question. I return None.
61 This is a wrapper around RIStorageServer.advise_corrupt_share(),
62 which is tied to a specific share, and therefore does not need the
63 extra share-identifying arguments. Please see that method for full
67 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
68 # elements are (offset, length, operator, specimen)
69 # operator is one of "lt, le, eq, ne, ge, gt"
70 # nop always passes and is used to fetch data while writing.
71 # you should use length==len(specimen) for everything except nop
72 DataVector = ListOf(TupleOf(Offset, ShareData))
73 # (offset, data). This limits us to 30 writes of 1MiB each per call
74 TestAndWriteVectorsForShares = DictOf(int,
77 ChoiceOf(None, Offset), # new_length
79 ReadVector = ListOf(TupleOf(Offset, ReadSize))
80 ReadData = ListOf(ShareData)
81 # returns data[offset:offset+length] for each element of TestVector
83 class RIStorageServer(RemoteInterface):
84 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
88 Return a dictionary of version information.
90 return DictOf(str, Any())
92 def allocate_buckets(storage_index=StorageIndex,
93 renew_secret=LeaseRenewSecret,
94 cancel_secret=LeaseCancelSecret,
95 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
96 allocated_size=Offset, canary=Referenceable):
98 @param storage_index: the index of the bucket to be created or
100 @param sharenums: these are the share numbers (probably between 0 and
101 99) that the sender is proposing to store on this
103 @param renew_secret: This is the secret used to protect bucket refresh
104 This secret is generated by the client and
105 stored for later comparison by the server. Each
106 server is given a different secret.
107 @param cancel_secret: Like renew_secret, but protects bucket decref.
108 @param canary: If the canary is lost before close(), the bucket is
110 @return: tuple of (alreadygot, allocated), where alreadygot is what we
111 already have and allocated is what we hereby agree to accept.
112 New leases are added for shares in both lists.
114 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
115 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
117 def add_lease(storage_index=StorageIndex,
118 renew_secret=LeaseRenewSecret,
119 cancel_secret=LeaseCancelSecret):
121 Add a new lease on the given bucket. If the renew_secret matches an
122 existing lease, that lease will be renewed instead. If there is no
123 bucket for the given storage_index, return silently. (note that in
124 tahoe-1.3.0 and earlier, IndexError was raised if there was no
127 return Any() # returns None now, but future versions might change
129 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
131 Renew the lease on a given bucket, resetting the timer to 31 days.
132 Some networks will use this, some will not. If there is no bucket for
133 the given storage_index, IndexError will be raised.
135 For mutable shares, if the given renew_secret does not match an
136 existing lease, IndexError will be raised with a note listing the
137 server-nodeids on the existing leases, so leases on migrated shares
138 can be renewed or cancelled. For immutable shares, IndexError
139 (without the note) will be raised.
143 def get_buckets(storage_index=StorageIndex):
144 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
148 def slot_readv(storage_index=StorageIndex,
149 shares=ListOf(int), readv=ReadVector):
150 """Read a vector from the numbered shares associated with the given
151 storage index. An empty shares list means to return data from all
152 known shares. Returns a dictionary with one key per share."""
153 return DictOf(int, ReadData) # shnum -> results
155 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
156 secrets=TupleOf(WriteEnablerSecret,
159 tw_vectors=TestAndWriteVectorsForShares,
162 """General-purpose test-and-set operation for mutable slots. Perform
163 a bunch of comparisons against the existing shares. If they all pass,
164 then apply a bunch of write vectors to those shares. Then use the
165 read vectors to extract data from all the shares and return the data.
167 This method is, um, large. The goal is to allow clients to update all
168 the shares associated with a mutable file in a single round trip.
170 @param storage_index: the index of the bucket to be created or
172 @param write_enabler: a secret that is stored along with the slot.
173 Writes are accepted from any caller who can
174 present the matching secret. A different secret
175 should be used for each slot*server pair.
176 @param renew_secret: This is the secret used to protect bucket refresh
177 This secret is generated by the client and
178 stored for later comparison by the server. Each
179 server is given a different secret.
180 @param cancel_secret: Like renew_secret, but protects bucket decref.
182 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
183 cancel_secret). The first is required to perform any write. The
184 latter two are used when allocating new shares. To simply acquire a
185 new lease on existing shares, use an empty testv and an empty writev.
187 Each share can have a separate test vector (i.e. a list of
188 comparisons to perform). If all vectors for all shares pass, then all
189 writes for all shares are recorded. Each comparison is a 4-tuple of
190 (offset, length, operator, specimen), which effectively does a bool(
191 (read(offset, length)) OPERATOR specimen ) and only performs the
192 write if all these evaluate to True. Basic test-and-set uses 'eq'.
193 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
194 Write-if-same-or-newer uses 'le'.
196 Reads from the end of the container are truncated, and missing shares
197 behave like empty ones, so to assert that a share doesn't exist (for
198 use when creating a new share), use (0, 1, 'eq', '').
200 The write vector will be applied to the given share, expanding it if
201 necessary. A write vector applied to a share number that did not
202 exist previously will cause that share to be created.
204 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
205 vector whose offset is beyond the end of the current data, the space
206 between the end of the current data and the beginning of the write
207 vector will be filled with zero bytes. In earlier versions the
208 contents of this space was unspecified (and might end up containing
209 secrets). Storage servers with the new zero-filling behavior will
210 advertise a true value for the 'fills-holes-with-zero-bytes' key
211 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
214 Each write vector is accompanied by a 'new_length' argument, which
215 can be used to truncate the data. If new_length is not None and it is
216 less than the current size of the data (after applying all write
217 vectors), then the data will be truncated to new_length. If
218 new_length==0, the share will be deleted.
220 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
221 enlarge the file by sending a number larger than the size of the data
222 after applying all write vectors. That behavior was not used, and as
223 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
226 If a storage client knows that the server supports zero-filling, for
227 example from the 'fills-holes-with-zero-bytes' key in its version
228 information, it can extend the file efficiently by writing a single
229 zero byte just before the new end-of-file. Otherwise it must
230 explicitly write zeroes to all bytes between the old and new
231 end-of-file. In any case it should avoid sending new_length larger
232 than the size of the data after applying all write vectors.
234 The read vector is used to extract data from all known shares,
235 *before* any writes have been applied. The same vector is used for
236 all shares. This captures the state that was tested by the test
239 This method returns two values: a boolean and a dict. The boolean is
240 True if the write vectors were applied, False if not. The dict is
241 keyed by share number, and each value contains a list of strings, one
242 for each element of the read vector.
244 If the write_enabler is wrong, this will raise BadWriteEnablerError.
245 To enable share migration (using update_write_enabler), the exception
246 will have the nodeid used for the old write enabler embedded in it,
247 in the following string::
249 The write enabler was recorded by nodeid '%s'.
251 Note that the nodeid here is encoded using the same base32 encoding
252 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
255 return TupleOf(bool, DictOf(int, ReadData))
257 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
258 shnum=int, reason=str):
259 """Clients who discover hash failures in shares that they have
260 downloaded from me will use this method to inform me about the
261 failures. I will record their concern so that my operator can
262 manually inspect the shares in question. I return None.
264 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
265 (binary) storage index string, and 'shnum' is the integer share
266 number. 'reason' is a human-readable explanation of the problem,
267 probably including some expected hash values and the computed ones
268 which did not match. Corruption advisories for mutable shares should
269 include a hash of the public key (the same value that appears in the
270 mutable-file verify-cap), since the current share format does not
274 class IStorageBucketWriter(Interface):
276 Objects of this kind live on the client side.
278 def put_block(segmentnum=int, data=ShareData):
279 """@param data: For most segments, this data will be 'blocksize'
280 bytes in length. The last segment might be shorter.
281 @return: a Deferred that fires (with None) when the operation completes
284 def put_plaintext_hashes(hashes=ListOf(Hash)):
286 @return: a Deferred that fires (with None) when the operation completes
289 def put_crypttext_hashes(hashes=ListOf(Hash)):
291 @return: a Deferred that fires (with None) when the operation completes
294 def put_block_hashes(blockhashes=ListOf(Hash)):
296 @return: a Deferred that fires (with None) when the operation completes
299 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
301 @return: a Deferred that fires (with None) when the operation completes
304 def put_uri_extension(data=URIExtensionData):
305 """This block of data contains integrity-checking information (hashes
306 of plaintext, crypttext, and shares), as well as encoding parameters
307 that are necessary to recover the data. This is a serialized dict
308 mapping strings to other strings. The hash of this data is kept in
309 the URI and verified before any of the data is used. All buckets for
310 a given file contain identical copies of this data.
312 The serialization format is specified with the following pseudocode:
313 for k in sorted(dict.keys()):
314 assert re.match(r'^[a-zA-Z_\-]+$', k)
315 write(k + ':' + netstring(dict[k]))
317 @return: a Deferred that fires (with None) when the operation completes
321 """Finish writing and close the bucket. The share is not finalized
322 until this method is called: if the uploading client disconnects
323 before calling close(), the partially-written share will be
326 @return: a Deferred that fires (with None) when the operation completes
329 class IStorageBucketReader(Interface):
331 def get_block_data(blocknum=int, blocksize=int, size=int):
332 """Most blocks will be the same size. The last block might be shorter
338 def get_crypttext_hashes():
340 @return: ListOf(Hash)
343 def get_block_hashes(at_least_these=SetOf(int)):
345 @return: ListOf(Hash)
348 def get_share_hashes(at_least_these=SetOf(int)):
350 @return: ListOf(TupleOf(int, Hash))
353 def get_uri_extension():
355 @return: URIExtensionData
358 class IStorageBroker(Interface):
359 def get_servers_for_psi(peer_selection_index):
361 @return: list of IServer instances
363 def get_connected_servers():
365 @return: frozenset of connected IServer instances
367 def get_known_servers():
369 @return: frozenset of IServer instances
371 def get_all_serverids():
373 @return: frozenset of serverid strings
375 def get_nickname_for_serverid(serverid):
377 @return: unicode nickname, or None
380 # methods moved from IntroducerClient, need review
381 def get_all_connections():
382 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
383 each active connection we've established to a remote service. This is
384 mostly useful for unit tests that need to wait until a certain number
385 of connections have been made."""
387 def get_all_connectors():
388 """Return a dict that maps from (nodeid, service_name) to a
389 RemoteServiceConnector instance for all services that we are actively
390 trying to connect to. Each RemoteServiceConnector has the following
393 service_name: the type of service provided, like 'storage'
394 announcement_time: when we first heard about this service
395 last_connect_time: when we last established a connection
396 last_loss_time: when we last lost a connection
398 version: the peer's version, from the most recent connection
399 oldest_supported: the peer's oldest supported version, same
401 rref: the RemoteReference, if connected, otherwise None
402 remote_host: the IAddress, if connected, otherwise None
404 This method is intended for monitoring interfaces, such as a web page
405 which describes connecting and connected peers.
408 def get_all_peerids():
409 """Return a frozenset of all peerids to whom we have a connection (to
410 one or more services) established. Mostly useful for unit tests."""
412 def get_all_connections_for(service_name):
413 """Return a frozenset of (nodeid, service_name, rref) tuples, one
414 for each active connection that provides the given SERVICE_NAME."""
416 def get_permuted_peers(service_name, key):
417 """Returns an ordered list of (peerid, rref) tuples, selecting from
418 the connections that provide SERVICE_NAME, using a hash-based
419 permutation keyed by KEY. This randomizes the service list in a
420 repeatable way, to distribute load over many peers.
424 class IMutableSlotWriter(Interface):
426 The interface for a writer around a mutable slot on a remote server.
428 def set_checkstring(checkstring, *args):
430 Set the checkstring that I will pass to the remote server when
433 @param checkstring A packed checkstring to use.
435 Note that implementations can differ in which semantics they
436 wish to support for set_checkstring -- they can, for example,
437 build the checkstring themselves from its constituents, or
441 def get_checkstring():
443 Get the checkstring that I think currently exists on the remote
447 def put_block(data, segnum, salt):
449 Add a block and salt to the share.
452 def put_encprivkey(encprivkey):
454 Add the encrypted private key to the share.
457 def put_blockhashes(blockhashes=list):
459 Add the block hash tree to the share.
462 def put_sharehashes(sharehashes=dict):
464 Add the share hash chain to the share.
469 Return the part of the share that needs to be signed.
472 def put_signature(signature):
474 Add the signature to the share.
477 def put_verification_key(verification_key):
479 Add the verification key to the share.
482 def finish_publishing():
484 Do anything necessary to finish writing the share to a remote
485 server. I require that no further publishing needs to take place
486 after this method has been called.
490 class IURI(Interface):
491 def init_from_string(uri):
492 """Accept a string (as created by my to_string() method) and populate
493 this instance with its data. I am not normally called directly,
494 please use the module-level uri.from_string() function to convert
495 arbitrary URI strings into IURI-providing instances."""
498 """Return False if this URI be used to modify the data. Return True
499 if this URI cannot be used to modify the data."""
502 """Return True if the data can be modified by *somebody* (perhaps
503 someone who has a more powerful URI than this one)."""
505 # TODO: rename to get_read_cap()
507 """Return another IURI instance, which represents a read-only form of
508 this one. If is_readonly() is True, this returns self."""
510 def get_verify_cap():
511 """Return an instance that provides IVerifierURI, which can be used
512 to check on the availability of the file or directory, without
513 providing enough capabilities to actually read or modify the
514 contents. This may return None if the file does not need checking or
515 verification (e.g. LIT URIs).
519 """Return a string of printable ASCII characters, suitable for
520 passing into init_from_string."""
522 class IVerifierURI(Interface, IURI):
523 def init_from_string(uri):
524 """Accept a string (as created by my to_string() method) and populate
525 this instance with its data. I am not normally called directly,
526 please use the module-level uri.from_string() function to convert
527 arbitrary URI strings into IURI-providing instances."""
530 """Return a string of printable ASCII characters, suitable for
531 passing into init_from_string."""
533 class IDirnodeURI(Interface):
534 """I am a URI which represents a dirnode."""
536 class IFileURI(Interface):
537 """I am a URI which represents a filenode."""
539 """Return the length (in bytes) of the file that I represent."""
541 class IImmutableFileURI(IFileURI):
544 class IMutableFileURI(Interface):
547 class IDirectoryURI(Interface):
550 class IReadonlyDirectoryURI(Interface):
553 class CapConstraintError(Exception):
554 """A constraint on a cap was violated."""
556 class MustBeDeepImmutableError(CapConstraintError):
557 """Mutable children cannot be added to an immutable directory.
558 Also, caps obtained from an immutable directory can trigger this error
559 if they are later found to refer to a mutable object and then used."""
561 class MustBeReadonlyError(CapConstraintError):
562 """Known write caps cannot be specified in a ro_uri field. Also,
563 caps obtained from a ro_uri field can trigger this error if they
564 are later found to be write caps and then used."""
566 class MustNotBeUnknownRWError(CapConstraintError):
567 """Cannot add an unknown child cap specified in a rw_uri field."""
570 class IReadable(Interface):
571 """I represent a readable object -- either an immutable file, or a
572 specific version of a mutable file.
576 """Return True if this reference provides mutable access to the given
577 file or directory (i.e. if you can modify it), or False if not. Note
578 that even if this reference is read-only, someone else may hold a
579 read-write reference to it.
581 For an IReadable returned by get_best_readable_version(), this will
582 always return True, but for instances of subinterfaces such as
583 IMutableFileVersion, it may return False."""
586 """Return True if this file or directory is mutable (by *somebody*,
587 not necessarily you), False if it is is immutable. Note that a file
588 might be mutable overall, but your reference to it might be
589 read-only. On the other hand, all references to an immutable file
590 will be read-only; there are no read-write references to an immutable
593 def get_storage_index():
594 """Return the storage index of the file."""
597 """Return the length (in bytes) of this readable object."""
599 def download_to_data():
600 """Download all of the file contents. I return a Deferred that fires
601 with the contents as a byte string."""
603 def read(consumer, offset=0, size=None):
604 """Download a portion (possibly all) of the file's contents, making
605 them available to the given IConsumer. Return a Deferred that fires
606 (with the consumer) when the consumer is unregistered (either because
607 the last byte has been given to it, or because the consumer threw an
608 exception during write(), possibly because it no longer wants to
609 receive data). The portion downloaded will start at 'offset' and
610 contain 'size' bytes (or the remainder of the file if size==None).
612 The consumer will be used in non-streaming mode: an IPullProducer
613 will be attached to it.
615 The consumer will not receive data right away: several network trips
616 must occur first. The order of events will be::
618 consumer.registerProducer(p, streaming)
619 (if streaming == False)::
620 consumer does p.resumeProducing()
622 consumer does p.resumeProducing()
623 consumer.write(data).. (repeat until all data is written)
624 consumer.unregisterProducer()
625 deferred.callback(consumer)
627 If a download error occurs, or an exception is raised by
628 consumer.registerProducer() or consumer.write(), I will call
629 consumer.unregisterProducer() and then deliver the exception via
630 deferred.errback(). To cancel the download, the consumer should call
631 p.stopProducing(), which will result in an exception being delivered
632 via deferred.errback().
634 See src/allmydata/util/consumer.py for an example of a simple
635 download-to-memory consumer.
639 class IWriteable(Interface):
641 I define methods that callers can use to update SDMF and MDMF
642 mutable files on a Tahoe-LAFS grid.
644 # XXX: For the moment, we have only this. It is possible that we
645 # want to move overwrite() and modify() in here too.
646 def update(data, offset):
648 I write the data from my data argument to the MDMF file,
649 starting at offset. I continue writing data until my data
650 argument is exhausted, appending data to the file as necessary.
652 # assert IMutableUploadable.providedBy(data)
653 # to append data: offset=node.get_size_of_best_version()
654 # do we want to support compacting MDMF?
655 # for an MDMF file, this can be done with O(data.get_size())
656 # memory. For an SDMF file, any modification takes
657 # O(node.get_size_of_best_version()).
660 class IMutableFileVersion(IReadable):
661 """I provide access to a particular version of a mutable file. The
662 access is read/write if I was obtained from a filenode derived from
663 a write cap, or read-only if the filenode was derived from a read cap.
666 def get_sequence_number():
667 """Return the sequence number of this version."""
670 """Return the IMutableFileServerMap instance that was used to create
675 """Return this filenode's writekey, or None if the node does not have
676 write-capability. This may be used to assist with data structures
677 that need to make certain data available only to writers, such as the
678 read-write child caps in dirnodes. The recommended process is to have
679 reader-visible data be submitted to the filenode in the clear (where
680 it will be encrypted by the filenode using the readkey), but encrypt
681 writer-visible data using this writekey.
684 # TODO: Can this be overwrite instead of replace?
685 def replace(new_contents):
686 """Replace the contents of the mutable file, provided that no other
687 node has published (or is attempting to publish, concurrently) a
688 newer version of the file than this one.
690 I will avoid modifying any share that is different than the version
691 given by get_sequence_number(). However, if another node is writing
692 to the file at the same time as me, I may manage to update some shares
693 while they update others. If I see any evidence of this, I will signal
694 UncoordinatedWriteError, and the file will be left in an inconsistent
695 state (possibly the version you provided, possibly the old version,
696 possibly somebody else's version, and possibly a mix of shares from
699 The recommended response to UncoordinatedWriteError is to either
700 return it to the caller (since they failed to coordinate their
701 writes), or to attempt some sort of recovery. It may be sufficient to
702 wait a random interval (with exponential backoff) and repeat your
703 operation. If I do not signal UncoordinatedWriteError, then I was
704 able to write the new version without incident.
706 I return a Deferred that fires (with a PublishStatus object) when the
707 update has completed.
710 def modify(modifier_cb):
711 """Modify the contents of the file, by downloading this version,
712 applying the modifier function (or bound method), then uploading
713 the new version. This will succeed as long as no other node
714 publishes a version between the download and the upload.
715 I return a Deferred that fires (with a PublishStatus object) when
716 the update is complete.
718 The modifier callable will be given three arguments: a string (with
719 the old contents), a 'first_time' boolean, and a servermap. As with
720 download_to_data(), the old contents will be from this version,
721 but the modifier can use the servermap to make other decisions
722 (such as refusing to apply the delta if there are multiple parallel
723 versions, or if there is evidence of a newer unrecoverable version).
724 'first_time' will be True the first time the modifier is called,
725 and False on any subsequent calls.
727 The callable should return a string with the new contents. The
728 callable must be prepared to be called multiple times, and must
729 examine the input string to see if the change that it wants to make
730 is already present in the old version. If it does not need to make
731 any changes, it can either return None, or return its input string.
733 If the modifier raises an exception, it will be returned in the
738 # The hierarchy looks like this:
745 class IFilesystemNode(Interface):
747 """Return the strongest 'cap instance' associated with this node.
748 (writecap for writeable-mutable files/directories, readcap for
749 immutable or readonly-mutable files/directories). To convert this
750 into a string, call .to_string() on the result."""
753 """Return a readonly cap instance for this node. For immutable or
754 readonly nodes, get_cap() and get_readcap() return the same thing."""
756 def get_repair_cap():
757 """Return an IURI instance that can be used to repair the file, or
758 None if this node cannot be repaired (either because it is not
759 distributed, like a LIT file, or because the node does not represent
760 sufficient authority to create a repair-cap, like a read-only RSA
761 mutable file node [which cannot create the correct write-enablers]).
764 def get_verify_cap():
765 """Return an IVerifierURI instance that represents the
766 'verifiy/refresh capability' for this node. The holder of this
767 capability will be able to renew the lease for this node, protecting
768 it from garbage-collection. They will also be able to ask a server if
769 it holds a share for the file or directory.
773 """Return the URI string corresponding to the strongest cap associated
774 with this node. If this node is read-only, the URI will only offer
775 read-only access. If this node is read-write, the URI will offer
778 If you have read-write access to a node and wish to share merely
779 read-only access with others, use get_readonly_uri().
783 """Return the URI string that can be used by others to get write
784 access to this node, if it is writeable. If this is a read-only node,
787 def get_readonly_uri():
788 """Return the URI string that can be used by others to get read-only
789 access to this node. The result is a read-only URI, regardless of
790 whether this node is read-only or read-write.
792 If you have merely read-only access to this node, get_readonly_uri()
793 will return the same thing as get_uri().
796 def get_storage_index():
797 """Return a string with the (binary) storage index in use on this
798 download. This may be None if there is no storage index (i.e. LIT
799 files and directories)."""
802 """Return True if this reference provides mutable access to the given
803 file or directory (i.e. if you can modify it), or False if not. Note
804 that even if this reference is read-only, someone else may hold a
805 read-write reference to it."""
808 """Return True if this file or directory is mutable (by *somebody*,
809 not necessarily you), False if it is is immutable. Note that a file
810 might be mutable overall, but your reference to it might be
811 read-only. On the other hand, all references to an immutable file
812 will be read-only; there are no read-write references to an immutable
817 """Return True if this is an unknown node."""
819 def is_allowed_in_immutable_directory():
820 """Return True if this node is allowed as a child of a deep-immutable
821 directory. This is true if either the node is of a known-immutable type,
822 or it is unknown and read-only.
826 """Raise any error associated with this node."""
828 # XXX: These may not be appropriate outside the context of an IReadable.
830 """Return the length (in bytes) of the data this node represents. For
831 directory nodes, I return the size of the backing store. I return
832 synchronously and do not consult the network, so for mutable objects,
833 I will return the most recently observed size for the object, or None
834 if I don't remember a size. Use get_current_size, which returns a
835 Deferred, if you want more up-to-date information."""
837 def get_current_size():
838 """I return a Deferred that fires with the length (in bytes) of the
839 data this node represents.
842 class IFileNode(IFilesystemNode):
843 """I am a node which represents a file: a sequence of bytes. I am not a
844 container, like IDirectoryNode."""
845 def get_best_readable_version():
846 """Return a Deferred that fires with an IReadable for the 'best'
847 available version of the file. The IReadable provides only read
848 access, even if this filenode was derived from a write cap.
850 For an immutable file, there is only one version. For a mutable
851 file, the 'best' version is the recoverable version with the
852 highest sequence number. If no uncoordinated writes have occurred,
853 and if enough shares are available, then this will be the most
854 recent version that has been uploaded. If no version is recoverable,
855 the Deferred will errback with an UnrecoverableFileError.
858 def download_best_version():
859 """Download the contents of the version that would be returned
860 by get_best_readable_version(). This is equivalent to calling
861 download_to_data() on the IReadable given by that method.
863 I return a Deferred that fires with a byte string when the file
864 has been fully downloaded. To support streaming download, use
865 the 'read' method of IReadable. If no version is recoverable,
866 the Deferred will errback with an UnrecoverableFileError.
869 def get_size_of_best_version():
870 """Find the size of the version that would be returned by
871 get_best_readable_version().
873 I return a Deferred that fires with an integer. If no version
874 is recoverable, the Deferred will errback with an
875 UnrecoverableFileError.
879 class IImmutableFileNode(IFileNode, IReadable):
880 """I am a node representing an immutable file. Immutable files have
884 class IMutableFileNode(IFileNode):
885 """I provide access to a 'mutable file', which retains its identity
886 regardless of what contents are put in it.
888 The consistency-vs-availability problem means that there might be
889 multiple versions of a file present in the grid, some of which might be
890 unrecoverable (i.e. have fewer than 'k' shares). These versions are
891 loosely ordered: each has a sequence number and a hash, and any version
892 with seqnum=N was uploaded by a node which has seen at least one version
895 The 'servermap' (an instance of IMutableFileServerMap) is used to
896 describe the versions that are known to be present in the grid, and which
897 servers are hosting their shares. It is used to represent the 'state of
898 the world', and is used for this purpose by my test-and-set operations.
899 Downloading the contents of the mutable file will also return a
900 servermap. Uploading a new version into the mutable file requires a
901 servermap as input, and the semantics of the replace operation is
902 'replace the file with my new version if it looks like nobody else has
903 changed the file since my previous download'. Because the file is
904 distributed, this is not a perfect test-and-set operation, but it will do
905 its best. If the replace process sees evidence of a simultaneous write,
906 it will signal an UncoordinatedWriteError, so that the caller can take
910 Most readers will want to use the 'best' current version of the file, and
911 should use my 'download_best_version()' method.
913 To unconditionally replace the file, callers should use overwrite(). This
914 is the mode that user-visible mutable files will probably use.
916 To apply some delta to the file, call modify() with a callable modifier
917 function that can apply the modification that you want to make. This is
918 the mode that dirnodes will use, since most directory modification
919 operations can be expressed in terms of deltas to the directory state.
922 Three methods are available for users who need to perform more complex
923 operations. The first is get_servermap(), which returns an up-to-date
924 servermap using a specified mode. The second is download_version(), which
925 downloads a specific version (not necessarily the 'best' one). The third
926 is 'upload', which accepts new contents and a servermap (which must have
927 been updated with MODE_WRITE). The upload method will attempt to apply
928 the new contents as long as no other node has modified the file since the
929 servermap was updated. This might be useful to a caller who wants to
930 merge multiple versions into a single new one.
932 Note that each time the servermap is updated, a specific 'mode' is used,
933 which determines how many peers are queried. To use a servermap for my
934 replace() method, that servermap must have been updated in MODE_WRITE.
935 These modes are defined in allmydata.mutable.common, and consist of
936 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
937 allmydata/mutable/servermap.py for details about the differences.
939 Mutable files are currently limited in size (about 3.5MB max) and can
940 only be retrieved and updated all-at-once, as a single big string. Future
941 versions of our mutable files will remove this restriction.
943 def get_best_mutable_version():
944 """Return a Deferred that fires with an IMutableFileVersion for
945 the 'best' available version of the file. The best version is
946 the recoverable version with the highest sequence number. If no
947 uncoordinated writes have occurred, and if enough shares are
948 available, then this will be the most recent version that has
951 If no version is recoverable, the Deferred will errback with an
952 UnrecoverableFileError.
955 def overwrite(new_contents):
956 """Unconditionally replace the contents of the mutable file with new
957 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
958 is only appropriate to use when the new contents of the file are
959 completely unrelated to the old ones, and you do not care about other
962 I return a Deferred that fires (with a PublishStatus object) when the
963 update has completed.
966 def modify(modifier_cb):
967 """Modify the contents of the file, by downloading the current
968 version, applying the modifier function (or bound method), then
969 uploading the new version. I return a Deferred that fires (with a
970 PublishStatus object) when the update is complete.
972 The modifier callable will be given three arguments: a string (with
973 the old contents), a 'first_time' boolean, and a servermap. As with
974 download_best_version(), the old contents will be from the best
975 recoverable version, but the modifier can use the servermap to make
976 other decisions (such as refusing to apply the delta if there are
977 multiple parallel versions, or if there is evidence of a newer
978 unrecoverable version). 'first_time' will be True the first time the
979 modifier is called, and False on any subsequent calls.
981 The callable should return a string with the new contents. The
982 callable must be prepared to be called multiple times, and must
983 examine the input string to see if the change that it wants to make
984 is already present in the old version. If it does not need to make
985 any changes, it can either return None, or return its input string.
987 If the modifier raises an exception, it will be returned in the
991 def get_servermap(mode):
992 """Return a Deferred that fires with an IMutableFileServerMap
993 instance, updated using the given mode.
996 def download_version(servermap, version):
997 """Download a specific version of the file, using the servermap
998 as a guide to where the shares are located.
1000 I return a Deferred that fires with the requested contents, or
1001 errbacks with UnrecoverableFileError. Note that a servermap which was
1002 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1003 all versions (those modes stop querying servers as soon as they can
1004 fulfil their goals), so you may want to use MODE_CHECK (which checks
1005 everything) to get increased visibility.
1008 def upload(new_contents, servermap):
1009 """Replace the contents of the file with new ones. This requires a
1010 servermap that was previously updated with MODE_WRITE.
1012 I attempt to provide test-and-set semantics, in that I will avoid
1013 modifying any share that is different than the version I saw in the
1014 servermap. However, if another node is writing to the file at the
1015 same time as me, I may manage to update some shares while they update
1016 others. If I see any evidence of this, I will signal
1017 UncoordinatedWriteError, and the file will be left in an inconsistent
1018 state (possibly the version you provided, possibly the old version,
1019 possibly somebody else's version, and possibly a mix of shares from
1022 The recommended response to UncoordinatedWriteError is to either
1023 return it to the caller (since they failed to coordinate their
1024 writes), or to attempt some sort of recovery. It may be sufficient to
1025 wait a random interval (with exponential backoff) and repeat your
1026 operation. If I do not signal UncoordinatedWriteError, then I was
1027 able to write the new version without incident.
1029 I return a Deferred that fires (with a PublishStatus object) when the
1030 publish has completed. I will update the servermap in-place with the
1031 location of all new shares.
1035 """Return this filenode's writekey, or None if the node does not have
1036 write-capability. This may be used to assist with data structures
1037 that need to make certain data available only to writers, such as the
1038 read-write child caps in dirnodes. The recommended process is to have
1039 reader-visible data be submitted to the filenode in the clear (where
1040 it will be encrypted by the filenode using the readkey), but encrypt
1041 writer-visible data using this writekey.
1045 """Returns the mutable file protocol version."""
1047 class NotEnoughSharesError(Exception):
1048 """Download was unable to get enough shares"""
1050 class NoSharesError(Exception):
1051 """Download was unable to get any shares at all."""
1053 class DownloadStopped(Exception):
1056 class UploadUnhappinessError(Exception):
1057 """Upload was unable to satisfy 'servers_of_happiness'"""
1059 class UnableToFetchCriticalDownloadDataError(Exception):
1060 """I was unable to fetch some piece of critical data which is supposed to
1061 be identically present in all shares."""
1063 class NoServersError(Exception):
1064 """Upload wasn't given any servers to work with, usually indicating a
1065 network or Introducer problem."""
1067 class ExistingChildError(Exception):
1068 """A directory node was asked to add or replace a child that already
1069 exists, and overwrite= was set to False."""
1071 class NoSuchChildError(Exception):
1072 """A directory node was asked to fetch a child which does not exist."""
1074 class ChildOfWrongTypeError(Exception):
1075 """An operation was attempted on a child of the wrong type (file or directory)."""
1077 class IDirectoryNode(IFilesystemNode):
1078 """I represent a filesystem node that is a container, with a
1079 name-to-child mapping, holding the tahoe equivalent of a directory. All
1080 child names are unicode strings, and all children are some sort of
1081 IFilesystemNode (a file, subdirectory, or unknown node).
1086 The dirnode ('1') URI returned by this method can be used in
1087 set_uri() on a different directory ('2') to 'mount' a reference to
1088 this directory ('1') under the other ('2'). This URI is just a
1089 string, so it can be passed around through email or other out-of-band
1093 def get_readonly_uri():
1095 The dirnode ('1') URI returned by this method can be used in
1096 set_uri() on a different directory ('2') to 'mount' a reference to
1097 this directory ('1') under the other ('2'). This URI is just a
1098 string, so it can be passed around through email or other out-of-band
1103 """I return a Deferred that fires with a dictionary mapping child
1104 name (a unicode string) to (node, metadata_dict) tuples, in which
1105 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1108 def has_child(name):
1109 """I return a Deferred that fires with a boolean, True if there
1110 exists a child of the given name, False if not. The child name must
1111 be a unicode string."""
1114 """I return a Deferred that fires with a specific named child node,
1115 which is an IFilesystemNode. The child name must be a unicode string.
1116 I raise NoSuchChildError if I do not have a child by that name."""
1118 def get_metadata_for(name):
1119 """I return a Deferred that fires with the metadata dictionary for
1120 a specific named child node. The child name must be a unicode string.
1121 This metadata is stored in the *edge*, not in the child, so it is
1122 attached to the parent dirnode rather than the child node.
1123 I raise NoSuchChildError if I do not have a child by that name."""
1125 def set_metadata_for(name, metadata):
1126 """I replace any existing metadata for the named child with the new
1127 metadata. The child name must be a unicode string. This metadata is
1128 stored in the *edge*, not in the child, so it is attached to the
1129 parent dirnode rather than the child node. I return a Deferred
1130 (that fires with this dirnode) when the operation is complete.
1131 I raise NoSuchChildError if I do not have a child by that name."""
1133 def get_child_at_path(path):
1134 """Transform a child path into an IFilesystemNode.
1136 I perform a recursive series of 'get' operations to find the named
1137 descendant node. I return a Deferred that fires with the node, or
1138 errbacks with NoSuchChildError if the node could not be found.
1140 The path can be either a single string (slash-separated) or a list of
1141 path-name elements. All elements must be unicode strings.
1144 def get_child_and_metadata_at_path(path):
1145 """Transform a child path into an IFilesystemNode and metadata.
1147 I am like get_child_at_path(), but my Deferred fires with a tuple of
1148 (node, metadata). The metadata comes from the last edge. If the path
1149 is empty, the metadata will be an empty dictionary.
1152 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1153 """I add a child (by writecap+readcap) at the specific name. I return
1154 a Deferred that fires when the operation finishes. If overwrite= is
1155 True, I will replace any existing child of the same name, otherwise
1156 an existing child will cause me to return ExistingChildError. The
1157 child name must be a unicode string.
1159 The child caps could be for a file, or for a directory. If you have
1160 both the writecap and readcap, you should provide both arguments.
1161 If you have only one cap and don't know whether it is read-only,
1162 provide it as the writecap argument and leave the readcap as None.
1163 If you have only one cap that is known to be read-only, provide it
1164 as the readcap argument and leave the writecap as None.
1165 The filecaps are typically obtained from an IFilesystemNode with
1166 get_uri() and get_readonly_uri().
1168 If metadata= is provided, I will use it as the metadata for the named
1169 edge. This will replace any existing metadata. If metadata= is left
1170 as the default value of None, I will set ['mtime'] to the current
1171 time, and I will set ['ctime'] to the current time if there was not
1172 already a child by this name present. This roughly matches the
1173 ctime/mtime semantics of traditional filesystems. See the
1174 "About the metadata" section of webapi.txt for futher information.
1176 If this directory node is read-only, the Deferred will errback with a
1177 NotWriteableError."""
1179 def set_children(entries, overwrite=True):
1180 """Add multiple children (by writecap+readcap) to a directory node.
1181 Takes a dictionary, with childname as keys and (writecap, readcap)
1182 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1183 a Deferred that fires (with this dirnode) when the operation
1184 finishes. This is equivalent to calling set_uri() multiple times, but
1185 is much more efficient. All child names must be unicode strings.
1188 def set_node(name, child, metadata=None, overwrite=True):
1189 """I add a child at the specific name. I return a Deferred that fires
1190 when the operation finishes. This Deferred will fire with the child
1191 node that was just added. I will replace any existing child of the
1192 same name. The child name must be a unicode string. The 'child'
1193 instance must be an instance providing IFilesystemNode.
1195 If metadata= is provided, I will use it as the metadata for the named
1196 edge. This will replace any existing metadata. If metadata= is left
1197 as the default value of None, I will set ['mtime'] to the current
1198 time, and I will set ['ctime'] to the current time if there was not
1199 already a child by this name present. This roughly matches the
1200 ctime/mtime semantics of traditional filesystems. See the
1201 "About the metadata" section of webapi.txt for futher information.
1203 If this directory node is read-only, the Deferred will errback with a
1204 NotWriteableError."""
1206 def set_nodes(entries, overwrite=True):
1207 """Add multiple children to a directory node. Takes a dict mapping
1208 unicode childname to (child_node, metdata) tuples. If metdata=None,
1209 the original metadata is left unmodified. Returns a Deferred that
1210 fires (with this dirnode) when the operation finishes. This is
1211 equivalent to calling set_node() multiple times, but is much more
1214 def add_file(name, uploadable, metadata=None, overwrite=True):
1215 """I upload a file (using the given IUploadable), then attach the
1216 resulting ImmutableFileNode to the directory at the given name. I set
1217 metadata the same way as set_uri and set_node. The child name must be
1220 I return a Deferred that fires (with the IFileNode of the uploaded
1221 file) when the operation completes."""
1223 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1224 """I remove the child at the specific name. I return a Deferred that
1225 fires when the operation finishes. The child name must be a unicode
1226 string. If must_exist is True and I do not have a child by that name,
1227 I raise NoSuchChildError. If must_be_directory is True and the child
1228 is a file, or if must_be_file is True and the child is a directory,
1229 I raise ChildOfWrongTypeError."""
1231 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1232 """I create and attach a directory at the given name. The new
1233 directory can be empty, or it can be populated with children
1234 according to 'initial_children', which takes a dictionary in the same
1235 format as set_nodes (i.e. mapping unicode child name to (childnode,
1236 metadata) tuples). The child name must be a unicode string. I return
1237 a Deferred that fires (with the new directory node) when the
1238 operation finishes."""
1240 def move_child_to(current_child_name, new_parent, new_child_name=None,
1242 """I take one of my children and move them to a new parent. The child
1243 is referenced by name. On the new parent, the child will live under
1244 'new_child_name', which defaults to 'current_child_name'. TODO: what
1245 should we do about metadata? I return a Deferred that fires when the
1246 operation finishes. The child name must be a unicode string. I raise
1247 NoSuchChildError if I do not have a child by that name."""
1249 def build_manifest():
1250 """I generate a table of everything reachable from this directory.
1251 I also compute deep-stats as described below.
1253 I return a Monitor. The Monitor's results will be a dictionary with
1256 res['manifest']: a list of (path, cap) tuples for all nodes
1257 (directories and files) reachable from this one.
1258 'path' will be a tuple of unicode strings. The
1259 origin dirnode will be represented by an empty path
1261 res['verifycaps']: a list of (printable) verifycap strings, one for
1262 each reachable non-LIT node. This is a set:
1263 it will contain no duplicates.
1264 res['storage-index']: a list of (base32) storage index strings,
1265 one for each reachable non-LIT node. This is
1266 a set: it will contain no duplicates.
1267 res['stats']: a dictionary, the same that is generated by
1268 start_deep_stats() below.
1270 The Monitor will also have an .origin_si attribute with the (binary)
1271 storage index of the starting point.
1274 def start_deep_stats():
1275 """Return a Monitor, examining all nodes (directories and files)
1276 reachable from this one. The Monitor's results will be a dictionary
1277 with the following keys::
1279 count-immutable-files: count of how many CHK files are in the set
1280 count-mutable-files: same, for mutable files (does not include
1282 count-literal-files: same, for LIT files
1283 count-files: sum of the above three
1285 count-directories: count of directories
1287 size-immutable-files: total bytes for all CHK files in the set
1288 size-mutable-files (TODO): same, for current version of all mutable
1289 files, does not include directories
1290 size-literal-files: same, for LIT files
1291 size-directories: size of mutable files used by directories
1293 largest-directory: number of bytes in the largest directory
1294 largest-directory-children: number of children in the largest
1296 largest-immutable-file: number of bytes in the largest CHK file
1298 size-mutable-files is not yet implemented, because it would involve
1299 even more queries than deep_stats does.
1301 The Monitor will also have an .origin_si attribute with the (binary)
1302 storage index of the starting point.
1304 This operation will visit every directory node underneath this one,
1305 and can take a long time to run. On a typical workstation with good
1306 bandwidth, this can examine roughly 15 directories per second (and
1307 takes several minutes of 100% CPU for ~1700 directories).
1310 class ICodecEncoder(Interface):
1311 def set_params(data_size, required_shares, max_shares):
1312 """Set up the parameters of this encoder.
1314 This prepares the encoder to perform an operation that converts a
1315 single block of data into a number of shares, such that a future
1316 ICodecDecoder can use a subset of these shares to recover the
1317 original data. This operation is invoked by calling encode(). Once
1318 the encoding parameters are set up, the encode operation can be
1319 invoked multiple times.
1321 set_params() prepares the encoder to accept blocks of input data that
1322 are exactly 'data_size' bytes in length. The encoder will be prepared
1323 to produce 'max_shares' shares for each encode() operation (although
1324 see the 'desired_share_ids' to use less CPU). The encoding math will
1325 be chosen such that the decoder can get by with as few as
1326 'required_shares' of these shares and still reproduce the original
1327 data. For example, set_params(1000, 5, 5) offers no redundancy at
1328 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1330 Numerical Restrictions: 'data_size' is required to be an integral
1331 multiple of 'required_shares'. In general, the caller should choose
1332 required_shares and max_shares based upon their reliability
1333 requirements and the number of peers available (the total storage
1334 space used is roughly equal to max_shares*data_size/required_shares),
1335 then choose data_size to achieve the memory footprint desired (larger
1336 data_size means more efficient operation, smaller data_size means
1337 smaller memory footprint).
1339 In addition, 'max_shares' must be equal to or greater than
1340 'required_shares'. Of course, setting them to be equal causes
1341 encode() to degenerate into a particularly slow form of the 'split'
1344 See encode() for more details about how these parameters are used.
1346 set_params() must be called before any other ICodecEncoder methods
1351 """Return the 3-tuple of data_size, required_shares, max_shares"""
1353 def get_encoder_type():
1354 """Return a short string that describes the type of this encoder.
1356 There is required to be a global table of encoder classes. This method
1357 returns an index into this table; the value at this index is an
1358 encoder class, and this encoder is an instance of that class.
1361 def get_block_size():
1362 """Return the length of the shares that encode() will produce.
1365 def encode_proposal(data, desired_share_ids=None):
1366 """Encode some data.
1368 'data' must be a string (or other buffer object), and len(data) must
1369 be equal to the 'data_size' value passed earlier to set_params().
1371 This will return a Deferred that will fire with two lists. The first
1372 is a list of shares, each of which is a string (or other buffer
1373 object) such that len(share) is the same as what get_share_size()
1374 returned earlier. The second is a list of shareids, in which each is
1375 an integer. The lengths of the two lists will always be equal to each
1376 other. The user should take care to keep each share closely
1377 associated with its shareid, as one is useless without the other.
1379 The length of this output list will normally be the same as the value
1380 provided to the 'max_shares' parameter of set_params(). This may be
1381 different if 'desired_share_ids' is provided.
1383 'desired_share_ids', if provided, is required to be a sequence of
1384 ints, each of which is required to be >= 0 and < max_shares. If not
1385 provided, encode() will produce 'max_shares' shares, as if
1386 'desired_share_ids' were set to range(max_shares). You might use this
1387 if you initially thought you were going to use 10 peers, started
1388 encoding, and then two of the peers dropped out: you could use
1389 desired_share_ids= to skip the work (both memory and CPU) of
1390 producing shares for the peers which are no longer available.
1394 def encode(inshares, desired_share_ids=None):
1395 """Encode some data. This may be called multiple times. Each call is
1398 inshares is a sequence of length required_shares, containing buffers
1399 (i.e. strings), where each buffer contains the next contiguous
1400 non-overlapping segment of the input data. Each buffer is required to
1401 be the same length, and the sum of the lengths of the buffers is
1402 required to be exactly the data_size promised by set_params(). (This
1403 implies that the data has to be padded before being passed to
1404 encode(), unless of course it already happens to be an even multiple
1405 of required_shares in length.)
1407 Note: the requirement to break up your data into
1408 'required_shares' chunks of exactly the right length before
1409 calling encode() is surprising from point of view of a user
1410 who doesn't know how FEC works. It feels like an
1411 implementation detail that has leaked outside the abstraction
1412 barrier. Is there a use case in which the data to be encoded
1413 might already be available in pre-segmented chunks, such that
1414 it is faster or less work to make encode() take a list rather
1415 than splitting a single string?
1417 Yes, there is: suppose you are uploading a file with K=64,
1418 N=128, segsize=262,144. Then each in-share will be of size
1419 4096. If you use this .encode() API then your code could first
1420 read each successive 4096-byte chunk from the file and store
1421 each one in a Python string and store each such Python string
1422 in a Python list. Then you could call .encode(), passing that
1423 list as "inshares". The encoder would generate the other 64
1424 "secondary shares" and return to you a new list containing
1425 references to the same 64 Python strings that you passed in
1426 (as the primary shares) plus references to the new 64 Python
1429 (You could even imagine that your code could use readv() so
1430 that the operating system can arrange to get all of those
1431 bytes copied from the file into the Python list of Python
1432 strings as efficiently as possible instead of having a loop
1433 written in C or in Python to copy the next part of the file
1434 into the next string.)
1436 On the other hand if you instead use the .encode_proposal()
1437 API (above), then your code can first read in all of the
1438 262,144 bytes of the segment from the file into a Python
1439 string, then call .encode_proposal() passing the segment data
1440 as the "data" argument. The encoder would basically first
1441 split the "data" argument into a list of 64 in-shares of 4096
1442 byte each, and then do the same thing that .encode() does. So
1443 this would result in a little bit more copying of data and a
1444 little bit higher of a "maximum memory usage" during the
1445 process, although it might or might not make a practical
1446 difference for our current use cases.
1448 Note that "inshares" is a strange name for the parameter if
1449 you think of the parameter as being just for feeding in data
1450 to the codec. It makes more sense if you think of the result
1451 of this encoding as being the set of shares from inshares plus
1452 an extra set of "secondary shares" (or "check shares"). It is
1453 a surprising name! If the API is going to be surprising then
1454 the name should be surprising. If we switch to
1455 encode_proposal() above then we should also switch to an
1458 'desired_share_ids', if provided, is required to be a sequence of
1459 ints, each of which is required to be >= 0 and < max_shares. If not
1460 provided, encode() will produce 'max_shares' shares, as if
1461 'desired_share_ids' were set to range(max_shares). You might use this
1462 if you initially thought you were going to use 10 peers, started
1463 encoding, and then two of the peers dropped out: you could use
1464 desired_share_ids= to skip the work (both memory and CPU) of
1465 producing shares for the peers which are no longer available.
1467 For each call, encode() will return a Deferred that fires with two
1468 lists, one containing shares and the other containing the shareids.
1469 The get_share_size() method can be used to determine the length of
1470 the share strings returned by encode(). Each shareid is a small
1471 integer, exactly as passed into 'desired_share_ids' (or
1472 range(max_shares), if desired_share_ids was not provided).
1474 The shares and their corresponding shareids are required to be kept
1475 together during storage and retrieval. Specifically, the share data is
1476 useless by itself: the decoder needs to be told which share is which
1477 by providing it with both the shareid and the actual share data.
1479 This function will allocate an amount of memory roughly equal to::
1481 (max_shares - required_shares) * get_share_size()
1483 When combined with the memory that the caller must allocate to
1484 provide the input data, this leads to a memory footprint roughly
1485 equal to the size of the resulting encoded shares (i.e. the expansion
1486 factor times the size of the input segment).
1491 # returning a list of (shareidN,shareN) tuples instead of a pair of
1492 # lists (shareids..,shares..). Brian thought the tuples would
1493 # encourage users to keep the share and shareid together throughout
1494 # later processing, Zooko pointed out that the code to iterate
1495 # through two lists is not really more complicated than using a list
1496 # of tuples and there's also a performance improvement
1498 # having 'data_size' not required to be an integral multiple of
1499 # 'required_shares'. Doing this would require encode() to perform
1500 # padding internally, and we'd prefer to have any padding be done
1501 # explicitly by the caller. Yes, it is an abstraction leak, but
1502 # hopefully not an onerous one.
1505 class ICodecDecoder(Interface):
1506 def set_params(data_size, required_shares, max_shares):
1507 """Set the params. They have to be exactly the same ones that were
1508 used for encoding."""
1510 def get_needed_shares():
1511 """Return the number of shares needed to reconstruct the data.
1512 set_params() is required to be called before this."""
1514 def decode(some_shares, their_shareids):
1515 """Decode a partial list of shares into data.
1517 'some_shares' is required to be a sequence of buffers of sharedata, a
1518 subset of the shares returned by ICodecEncode.encode(). Each share is
1519 required to be of the same length. The i'th element of their_shareids
1520 is required to be the shareid of the i'th buffer in some_shares.
1522 This returns a Deferred which fires with a sequence of buffers. This
1523 sequence will contain all of the segments of the original data, in
1524 order. The sum of the lengths of all of the buffers will be the
1525 'data_size' value passed into the original ICodecEncode.set_params()
1526 call. To get back the single original input block of data, use
1527 ''.join(output_buffers), or you may wish to simply write them in
1528 order to an output file.
1530 Note that some of the elements in the result sequence may be
1531 references to the elements of the some_shares input sequence. In
1532 particular, this means that if those share objects are mutable (e.g.
1533 arrays) and if they are changed, then both the input (the
1534 'some_shares' parameter) and the output (the value given when the
1535 deferred is triggered) will change.
1537 The length of 'some_shares' is required to be exactly the value of
1538 'required_shares' passed into the original ICodecEncode.set_params()
1542 class IEncoder(Interface):
1543 """I take an object that provides IEncryptedUploadable, which provides
1544 encrypted data, and a list of shareholders. I then encode, hash, and
1545 deliver shares to those shareholders. I will compute all the necessary
1546 Merkle hash trees that are necessary to validate the crypttext that
1547 eventually comes back from the shareholders. I provide the URI Extension
1548 Block Hash, and the encoding parameters, both of which must be included
1551 I do not choose shareholders, that is left to the IUploader. I must be
1552 given a dict of RemoteReferences to storage buckets that are ready and
1553 willing to receive data.
1557 """Specify the number of bytes that will be encoded. This must be
1558 peformed before get_serialized_params() can be called.
1560 def set_params(params):
1561 """Override the default encoding parameters. 'params' is a tuple of
1562 (k,d,n), where 'k' is the number of required shares, 'd' is the
1563 servers_of_happiness, and 'n' is the total number of shares that will
1566 Encoding parameters can be set in three ways. 1: The Encoder class
1567 provides defaults (3/7/10). 2: the Encoder can be constructed with
1568 an 'options' dictionary, in which the
1569 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1570 set_params((k,d,n)) can be called.
1572 If you intend to use set_params(), you must call it before
1573 get_share_size or get_param are called.
1576 def set_encrypted_uploadable(u):
1577 """Provide a source of encrypted upload data. 'u' must implement
1578 IEncryptedUploadable.
1580 When this is called, the IEncryptedUploadable will be queried for its
1581 length and the storage_index that should be used.
1583 This returns a Deferred that fires with this Encoder instance.
1585 This must be performed before start() can be called.
1588 def get_param(name):
1589 """Return an encoding parameter, by name.
1591 'storage_index': return a string with the (16-byte truncated SHA-256
1592 hash) storage index to which these shares should be
1595 'share_counts': return a tuple describing how many shares are used:
1596 (needed_shares, servers_of_happiness, total_shares)
1598 'num_segments': return an int with the number of segments that
1601 'segment_size': return an int with the size of each segment.
1603 'block_size': return the size of the individual blocks that will
1604 be delivered to a shareholder's put_block() method. By
1605 knowing this, the shareholder will be able to keep all
1606 blocks in a single file and still provide random access
1607 when reading them. # TODO: can we avoid exposing this?
1609 'share_size': an int with the size of the data that will be stored
1610 on each shareholder. This is aggregate amount of data
1611 that will be sent to the shareholder, summed over all
1612 the put_block() calls I will ever make. It is useful to
1613 determine this size before asking potential
1614 shareholders whether they will grant a lease or not,
1615 since their answers will depend upon how much space we
1616 need. TODO: this might also include some amount of
1617 overhead, like the size of all the hashes. We need to
1618 decide whether this is useful or not.
1620 'serialized_params': a string with a concise description of the
1621 codec name and its parameters. This may be passed
1622 into the IUploadable to let it make sure that
1623 the same file encoded with different parameters
1624 will result in different storage indexes.
1626 Once this is called, set_size() and set_params() may not be called.
1629 def set_shareholders(shareholders, servermap):
1630 """Tell the encoder where to put the encoded shares. 'shareholders'
1631 must be a dictionary that maps share number (an integer ranging from
1632 0 to n-1) to an instance that provides IStorageBucketWriter.
1633 'servermap' is a dictionary that maps share number (as defined above)
1634 to a set of peerids. This must be performed before start() can be
1638 """Begin the encode/upload process. This involves reading encrypted
1639 data from the IEncryptedUploadable, encoding it, uploading the shares
1640 to the shareholders, then sending the hash trees.
1642 set_encrypted_uploadable() and set_shareholders() must be called
1643 before this can be invoked.
1645 This returns a Deferred that fires with a verify cap when the upload
1646 process is complete. The verifycap, plus the encryption key, is
1647 sufficient to construct the read cap.
1650 class IDecoder(Interface):
1651 """I take a list of shareholders and some setup information, then
1652 download, validate, decode, and decrypt data from them, writing the
1653 results to an output file.
1655 I do not locate the shareholders, that is left to the IDownloader. I must
1656 be given a dict of RemoteReferences to storage buckets that are ready to
1661 """I take a file-like object (providing write and close) to which all
1662 the plaintext data will be written.
1664 TODO: producer/consumer . Maybe write() should return a Deferred that
1665 indicates when it will accept more data? But probably having the
1666 IDecoder be a producer is easier to glue to IConsumer pieces.
1669 def set_shareholders(shareholders):
1670 """I take a dictionary that maps share identifiers (small integers)
1671 to RemoteReferences that provide RIBucketReader. This must be called
1675 """I start the download. This process involves retrieving data and
1676 hash chains from the shareholders, using the hashes to validate the
1677 data, decoding the shares into segments, decrypting the segments,
1678 then writing the resulting plaintext to the output file.
1680 I return a Deferred that will fire (with self) when the download is
1684 class IDownloadTarget(Interface):
1685 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1686 # will register itself as a producer. This allows the target to invoke
1687 # downloader.pauseProducing, resumeProducing, and stopProducing.
1689 """Called before any calls to write() or close(). If an error
1690 occurs before any data is available, fail() may be called without
1691 a previous call to open().
1693 'size' is the length of the file being downloaded, in bytes."""
1696 """Output some data to the target."""
1698 """Inform the target that there is no more data to be written."""
1700 """fail() is called to indicate that the download has failed. 'why'
1701 is a Failure object indicating what went wrong. No further methods
1702 will be invoked on the IDownloadTarget after fail()."""
1703 def register_canceller(cb):
1704 """The CiphertextDownloader uses this to register a no-argument function
1705 that the target can call to cancel the download. Once this canceller
1706 is invoked, no further calls to write() or close() will be made."""
1708 """When the CiphertextDownloader is done, this finish() function will be
1709 called. Whatever it returns will be returned to the invoker of
1710 Downloader.download.
1713 class IDownloader(Interface):
1714 def download(uri, target):
1715 """Perform a CHK download, sending the data to the given target.
1716 'target' must provide IDownloadTarget.
1718 Returns a Deferred that fires (with the results of target.finish)
1719 when the download is finished, or errbacks if something went wrong."""
1721 class IEncryptedUploadable(Interface):
1722 def set_upload_status(upload_status):
1723 """Provide an IUploadStatus object that should be filled with status
1724 information. The IEncryptedUploadable is responsible for setting
1725 key-determination progress ('chk'), size, storage_index, and
1726 ciphertext-fetch progress. It may delegate some of this
1727 responsibility to others, in particular to the IUploadable."""
1730 """This behaves just like IUploadable.get_size()."""
1732 def get_all_encoding_parameters():
1733 """Return a Deferred that fires with a tuple of
1734 (k,happy,n,segment_size). The segment_size will be used as-is, and
1735 must match the following constraints: it must be a multiple of k, and
1736 it shouldn't be unreasonably larger than the file size (if
1737 segment_size is larger than filesize, the difference must be stored
1740 This usually passes through to the IUploadable method of the same
1743 The encoder strictly obeys the values returned by this method. To
1744 make an upload use non-default encoding parameters, you must arrange
1745 to control the values that this method returns.
1748 def get_storage_index():
1749 """Return a Deferred that fires with a 16-byte storage index.
1752 def read_encrypted(length, hash_only):
1753 """This behaves just like IUploadable.read(), but returns crypttext
1754 instead of plaintext. If hash_only is True, then this discards the
1755 data (and returns an empty list); this improves efficiency when
1756 resuming an interrupted upload (where we need to compute the
1757 plaintext hashes, but don't need the redundant encrypted data)."""
1759 def get_plaintext_hashtree_leaves(first, last, num_segments):
1760 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1761 plaintext segments, i.e. get the tagged hashes of the given segments.
1762 The segment size is expected to be generated by the
1763 IEncryptedUploadable before any plaintext is read or ciphertext
1764 produced, so that the segment hashes can be generated with only a
1767 This returns a Deferred which fires with a sequence of hashes, using:
1769 tuple(segment_hashes[first:last])
1771 'num_segments' is used to assert that the number of segments that the
1772 IEncryptedUploadable handled matches the number of segments that the
1773 encoder was expecting.
1775 This method must not be called until the final byte has been read
1776 from read_encrypted(). Once this method is called, read_encrypted()
1777 can never be called again.
1780 def get_plaintext_hash():
1781 """OBSOLETE; Get the hash of the whole plaintext.
1783 This returns a Deferred which fires with a tagged SHA-256 hash of the
1784 whole plaintext, obtained from hashutil.plaintext_hash(data).
1788 """Just like IUploadable.close()."""
1790 class IUploadable(Interface):
1791 def set_upload_status(upload_status):
1792 """Provide an IUploadStatus object that should be filled with status
1793 information. The IUploadable is responsible for setting
1794 key-determination progress ('chk')."""
1796 def set_default_encoding_parameters(params):
1797 """Set the default encoding parameters, which must be a dict mapping
1798 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1799 'max_segment_size'. These might have an influence on the final
1800 encoding parameters returned by get_all_encoding_parameters(), if the
1801 Uploadable doesn't have more specific preferences.
1803 This call is optional: if it is not used, the Uploadable will use
1804 some built-in defaults. If used, this method must be called before
1805 any other IUploadable methods to have any effect.
1809 """Return a Deferred that will fire with the length of the data to be
1810 uploaded, in bytes. This will be called before the data is actually
1811 used, to compute encoding parameters.
1814 def get_all_encoding_parameters():
1815 """Return a Deferred that fires with a tuple of
1816 (k,happy,n,segment_size). The segment_size will be used as-is, and
1817 must match the following constraints: it must be a multiple of k, and
1818 it shouldn't be unreasonably larger than the file size (if
1819 segment_size is larger than filesize, the difference must be stored
1822 The relative values of k and n allow some IUploadables to request
1823 better redundancy than others (in exchange for consuming more space
1826 Larger values of segment_size reduce hash overhead, while smaller
1827 values reduce memory footprint and cause data to be delivered in
1828 smaller pieces (which may provide a smoother and more predictable
1829 download experience).
1831 The encoder strictly obeys the values returned by this method. To
1832 make an upload use non-default encoding parameters, you must arrange
1833 to control the values that this method returns. One way to influence
1834 them may be to call set_encoding_parameters() before calling
1835 get_all_encoding_parameters().
1838 def get_encryption_key():
1839 """Return a Deferred that fires with a 16-byte AES key. This key will
1840 be used to encrypt the data. The key will also be hashed to derive
1843 Uploadables which want to achieve convergence should hash their file
1844 contents and the serialized_encoding_parameters to form the key
1845 (which of course requires a full pass over the data). Uploadables can
1846 use the upload.ConvergentUploadMixin class to achieve this
1849 Uploadables which do not care about convergence (or do not wish to
1850 make multiple passes over the data) can simply return a
1851 strongly-random 16 byte string.
1853 get_encryption_key() may be called multiple times: the IUploadable is
1854 required to return the same value each time.
1858 """Return a Deferred that fires with a list of strings (perhaps with
1859 only a single element) which, when concatenated together, contain the
1860 next 'length' bytes of data. If EOF is near, this may provide fewer
1861 than 'length' bytes. The total number of bytes provided by read()
1862 before it signals EOF must equal the size provided by get_size().
1864 If the data must be acquired through multiple internal read
1865 operations, returning a list instead of a single string may help to
1866 reduce string copies. However, the length of the concatenated strings
1867 must equal the amount of data requested, unless EOF is encountered.
1868 Long reads, or short reads without EOF, are not allowed. read()
1869 should return the same amount of data as a local disk file read, just
1870 in a different shape and asynchronously.
1872 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1873 so a 10kB file means length=3kB, 100kB file means length=30kB,
1874 and >=1MB file means length=300kB.
1876 This method provides for a single full pass through the data. Later
1877 use cases may desire multiple passes or access to only parts of the
1878 data (such as a mutable file making small edits-in-place). This API
1879 will be expanded once those use cases are better understood.
1883 """The upload is finished, and whatever filehandle was in use may be
1887 class IMutableUploadable(Interface):
1889 I represent content that is due to be uploaded to a mutable filecap.
1891 # This is somewhat simpler than the IUploadable interface above
1892 # because mutable files do not need to be concerned with possibly
1893 # generating a CHK, nor with per-file keys. It is a subset of the
1894 # methods in IUploadable, though, so we could just as well implement
1895 # the mutable uploadables as IUploadables that don't happen to use
1896 # those methods (with the understanding that the unused methods will
1897 # never be called on such objects)
1900 Returns a Deferred that fires with the size of the content held
1906 Returns a list of strings which, when concatenated, are the next
1907 length bytes of the file, or fewer if there are fewer bytes
1908 between the current location and the end of the file.
1913 The process that used the Uploadable is finished using it, so
1914 the uploadable may be closed.
1917 class IUploadResults(Interface):
1918 """I am returned by upload() methods. I contain a number of public
1919 attributes which can be read to determine the results of the upload. Some
1920 of these are functional, some are timing information. All of these may be
1923 .file_size : the size of the file, in bytes
1924 .uri : the CHK read-cap for the file
1925 .ciphertext_fetched : how many bytes were fetched by the helper
1926 .sharemap: dict mapping share identifier to set of serverids
1927 (binary strings). This indicates which servers were given
1928 which shares. For immutable files, the shareid is an
1929 integer (the share number, from 0 to N-1). For mutable
1930 files, it is a string of the form 'seq%d-%s-sh%d',
1931 containing the sequence number, the roothash, and the
1933 .servermap : dict mapping server peerid to a set of share numbers
1934 .timings : dict of timing information, mapping name to seconds (float)
1935 total : total upload time, start to finish
1936 storage_index : time to compute the storage index
1937 peer_selection : time to decide which peers will be used
1938 contacting_helper : initial helper query to upload/no-upload decision
1939 existence_check : helper pre-upload existence check
1940 helper_total : initial helper query to helper finished pushing
1941 cumulative_fetch : helper waiting for ciphertext requests
1942 total_fetch : helper start to last ciphertext response
1943 cumulative_encoding : just time spent in zfec
1944 cumulative_sending : just time spent waiting for storage servers
1945 hashes_and_close : last segment push to shareholder close
1946 total_encode_and_push : first encode to shareholder close
1950 class IDownloadResults(Interface):
1951 """I am created internally by download() methods. I contain a number of
1952 public attributes which contain details about the download process.::
1954 .file_size : the size of the file, in bytes
1955 .servers_used : set of server peerids that were used during download
1956 .server_problems : dict mapping server peerid to a problem string. Only
1957 servers that had problems (bad hashes, disconnects)
1959 .servermap : dict mapping server peerid to a set of share numbers. Only
1960 servers that had any shares are listed here.
1961 .timings : dict of timing information, mapping name to seconds (float)
1962 peer_selection : time to ask servers about shares
1963 servers_peer_selection : dict of peerid to DYHB-query time
1964 uri_extension : time to fetch a copy of the URI extension block
1965 hashtrees : time to fetch the hash trees
1966 segments : time to fetch, decode, and deliver segments
1967 cumulative_fetch : time spent waiting for storage servers
1968 cumulative_decode : just time spent in zfec
1969 cumulative_decrypt : just time spent in decryption
1970 total : total download time, start to finish
1971 fetch_per_server : dict of peerid to list of per-segment fetch times
1975 class IUploader(Interface):
1976 def upload(uploadable):
1977 """Upload the file. 'uploadable' must impement IUploadable. This
1978 returns a Deferred which fires with an IUploadResults instance, from
1979 which the URI of the file can be obtained as results.uri ."""
1981 def upload_ssk(write_capability, new_version, uploadable):
1982 """TODO: how should this work?"""
1984 class ICheckable(Interface):
1985 def check(monitor, verify=False, add_lease=False):
1986 """Check up on my health, optionally repairing any problems.
1988 This returns a Deferred that fires with an instance that provides
1989 ICheckResults, or None if the object is non-distributed (i.e. LIT
1992 The monitor will be checked periodically to see if the operation has
1993 been cancelled. If so, no new queries will be sent, and the Deferred
1994 will fire (with a OperationCancelledError) immediately.
1996 Filenodes and dirnodes (which provide IFilesystemNode) are also
1997 checkable. Instances that represent verifier-caps will be checkable
1998 but not downloadable. Some objects (like LIT files) do not actually
1999 live in the grid, and their checkers return None (non-distributed
2000 files are always healthy).
2002 If verify=False, a relatively lightweight check will be performed: I
2003 will ask all servers if they have a share for me, and I will believe
2004 whatever they say. If there are at least N distinct shares on the
2005 grid, my results will indicate r.is_healthy()==True. This requires a
2006 roundtrip to each server, but does not transfer very much data, so
2007 the network bandwidth is fairly low.
2009 If verify=True, a more resource-intensive check will be performed:
2010 every share will be downloaded, and the hashes will be validated on
2011 every bit. I will ignore any shares that failed their hash checks. If
2012 there are at least N distinct valid shares on the grid, my results
2013 will indicate r.is_healthy()==True. This requires N/k times as much
2014 download bandwidth (and server disk IO) as a regular download. If a
2015 storage server is holding a corrupt share, or is experiencing memory
2016 failures during retrieval, or is malicious or buggy, then
2017 verification will detect the problem, but checking will not.
2019 If add_lease=True, I will ensure that an up-to-date lease is present
2020 on each share. The lease secrets will be derived from by node secret
2021 (in BASEDIR/private/secret), so either I will add a new lease to the
2022 share, or I will merely renew the lease that I already had. In a
2023 future version of the storage-server protocol (once Accounting has
2024 been implemented), there may be additional options here to define the
2025 kind of lease that is obtained (which account number to claim, etc).
2027 TODO: any problems seen during checking will be reported to the
2028 health-manager.furl, a centralized object which is responsible for
2029 figuring out why files are unhealthy so corrective action can be
2033 def check_and_repair(monitor, verify=False, add_lease=False):
2034 """Like check(), but if the file/directory is not healthy, attempt to
2037 Any non-healthy result will cause an immediate repair operation, to
2038 generate and upload new shares. After repair, the file will be as
2039 healthy as we can make it. Details about what sort of repair is done
2040 will be put in the check-and-repair results. The Deferred will not
2041 fire until the repair is complete.
2043 This returns a Deferred which fires with an instance of
2044 ICheckAndRepairResults."""
2046 class IDeepCheckable(Interface):
2047 def start_deep_check(verify=False, add_lease=False):
2048 """Check upon the health of me and everything I can reach.
2050 This is a recursive form of check(), useable only on dirnodes.
2052 I return a Monitor, with results that are an IDeepCheckResults
2055 TODO: If any of the directories I traverse are unrecoverable, the
2056 Monitor will report failure. If any of the files I check upon are
2057 unrecoverable, those problems will be reported in the
2058 IDeepCheckResults as usual, and the Monitor will not report a
2062 def start_deep_check_and_repair(verify=False, add_lease=False):
2063 """Check upon the health of me and everything I can reach. Repair
2064 anything that isn't healthy.
2066 This is a recursive form of check_and_repair(), useable only on
2069 I return a Monitor, with results that are an
2070 IDeepCheckAndRepairResults object.
2072 TODO: If any of the directories I traverse are unrecoverable, the
2073 Monitor will report failure. If any of the files I check upon are
2074 unrecoverable, those problems will be reported in the
2075 IDeepCheckResults as usual, and the Monitor will not report a
2079 class ICheckResults(Interface):
2080 """I contain the detailed results of a check/verify operation.
2083 def get_storage_index():
2084 """Return a string with the (binary) storage index."""
2085 def get_storage_index_string():
2086 """Return a string with the (printable) abbreviated storage index."""
2088 """Return the (string) URI of the object that was checked."""
2091 """Return a boolean, True if the file/dir is fully healthy, False if
2092 it is damaged in any way. Non-distributed LIT files always return
2095 def is_recoverable():
2096 """Return a boolean, True if the file/dir can be recovered, False if
2097 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2098 files always return True."""
2100 def needs_rebalancing():
2101 """Return a boolean, True if the file/dir's reliability could be
2102 improved by moving shares to new servers. Non-distributed LIT files
2103 always return False."""
2107 """Return a dictionary that describes the state of the file/dir. LIT
2108 files always return an empty dictionary. Normal files and directories
2109 return a dictionary with the following keys (note that these use
2110 binary strings rather than base32-encoded ones) (also note that for
2111 mutable files, these counts are for the 'best' version):
2113 count-shares-good: the number of distinct good shares that were found
2114 count-shares-needed: 'k', the number of shares required for recovery
2115 count-shares-expected: 'N', the number of total shares generated
2116 count-good-share-hosts: the number of distinct storage servers with
2117 good shares. If this number is less than
2118 count-shares-good, then some shares are
2119 doubled up, increasing the correlation of
2120 failures. This indicates that one or more
2121 shares should be moved to an otherwise unused
2122 server, if one is available.
2123 count-corrupt-shares: the number of shares with integrity failures
2124 list-corrupt-shares: a list of 'share locators', one for each share
2125 that was found to be corrupt. Each share
2126 locator is a list of (serverid, storage_index,
2128 count-incompatible-shares: the number of shares which are of a share
2129 format unknown to this checker
2130 list-incompatible-shares: a list of 'share locators', one for each
2131 share that was found to be of an unknown
2132 format. Each share locator is a list of
2133 (serverid, storage_index, sharenum).
2134 servers-responding: list of (binary) storage server identifiers,
2135 one for each server which responded to the share
2136 query (even if they said they didn't have
2137 shares, and even if they said they did have
2138 shares but then didn't send them when asked, or
2139 dropped the connection, or returned a Failure,
2140 and even if they said they did have shares and
2141 sent incorrect ones when asked)
2142 sharemap: dict mapping share identifier to list of serverids
2143 (binary strings). This indicates which servers are holding
2144 which shares. For immutable files, the shareid is an
2145 integer (the share number, from 0 to N-1). For mutable
2146 files, it is a string of the form 'seq%d-%s-sh%d',
2147 containing the sequence number, the roothash, and the
2150 The following keys are most relevant for mutable files, but immutable
2151 files will provide sensible values too::
2153 count-wrong-shares: the number of shares for versions other than the
2154 'best' one (which is defined as being the
2155 recoverable version with the highest sequence
2156 number, then the highest roothash). These are
2157 either leftover shares from an older version
2158 (perhaps on a server that was offline when an
2159 update occurred), shares from an unrecoverable
2160 newer version, or shares from an alternate
2161 current version that results from an
2162 uncoordinated write collision. For a healthy
2163 file, this will equal 0.
2165 count-recoverable-versions: the number of recoverable versions of
2166 the file. For a healthy file, this will
2169 count-unrecoverable-versions: the number of unrecoverable versions
2170 of the file. For a healthy file, this
2176 """Return a string with a brief (one-line) summary of the results."""
2179 """Return a list of strings with more detailed results."""
2181 class ICheckAndRepairResults(Interface):
2182 """I contain the detailed results of a check/verify/repair operation.
2184 The IFilesystemNode.check()/verify()/repair() methods all return
2185 instances that provide ICheckAndRepairResults.
2188 def get_storage_index():
2189 """Return a string with the (binary) storage index."""
2190 def get_storage_index_string():
2191 """Return a string with the (printable) abbreviated storage index."""
2192 def get_repair_attempted():
2193 """Return a boolean, True if a repair was attempted. We might not
2194 attempt to repair the file because it was healthy, or healthy enough
2195 (i.e. some shares were missing but not enough to exceed some
2196 threshold), or because we don't know how to repair this object."""
2197 def get_repair_successful():
2198 """Return a boolean, True if repair was attempted and the file/dir
2199 was fully healthy afterwards. False if no repair was attempted or if
2200 a repair attempt failed."""
2201 def get_pre_repair_results():
2202 """Return an ICheckResults instance that describes the state of the
2203 file/dir before any repair was attempted."""
2204 def get_post_repair_results():
2205 """Return an ICheckResults instance that describes the state of the
2206 file/dir after any repair was attempted. If no repair was attempted,
2207 the pre-repair and post-repair results will be identical."""
2210 class IDeepCheckResults(Interface):
2211 """I contain the results of a deep-check operation.
2213 This is returned by a call to ICheckable.deep_check().
2216 def get_root_storage_index_string():
2217 """Return the storage index (abbreviated human-readable string) of
2218 the first object checked."""
2220 """Return a dictionary with the following keys::
2222 count-objects-checked: count of how many objects were checked
2223 count-objects-healthy: how many of those objects were completely
2225 count-objects-unhealthy: how many were damaged in some way
2226 count-objects-unrecoverable: how many were unrecoverable
2227 count-corrupt-shares: how many shares were found to have
2228 corruption, summed over all objects
2232 def get_corrupt_shares():
2233 """Return a set of (serverid, storage_index, sharenum) for all shares
2234 that were found to be corrupt. Both serverid and storage_index are
2237 def get_all_results():
2238 """Return a dictionary mapping pathname (a tuple of strings, ready to
2239 be slash-joined) to an ICheckResults instance, one for each object
2240 that was checked."""
2242 def get_results_for_storage_index(storage_index):
2243 """Retrive the ICheckResults instance for the given (binary)
2244 storage index. Raises KeyError if there are no results for that
2248 """Return a dictionary with the same keys as
2249 IDirectoryNode.deep_stats()."""
2251 class IDeepCheckAndRepairResults(Interface):
2252 """I contain the results of a deep-check-and-repair operation.
2254 This is returned by a call to ICheckable.deep_check_and_repair().
2257 def get_root_storage_index_string():
2258 """Return the storage index (abbreviated human-readable string) of
2259 the first object checked."""
2261 """Return a dictionary with the following keys::
2263 count-objects-checked: count of how many objects were checked
2264 count-objects-healthy-pre-repair: how many of those objects were
2265 completely healthy (before any
2267 count-objects-unhealthy-pre-repair: how many were damaged in
2269 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2270 count-objects-healthy-post-repair: how many of those objects were
2271 completely healthy (after any
2273 count-objects-unhealthy-post-repair: how many were damaged in
2275 count-objects-unrecoverable-post-repair: how many were
2277 count-repairs-attempted: repairs were attempted on this many
2278 objects. The count-repairs- keys will
2279 always be provided, however unless
2280 repair=true is present, they will all
2282 count-repairs-successful: how many repairs resulted in healthy
2284 count-repairs-unsuccessful: how many repairs resulted did not
2285 results in completely healthy objects
2286 count-corrupt-shares-pre-repair: how many shares were found to
2287 have corruption, summed over all
2288 objects examined (before any
2290 count-corrupt-shares-post-repair: how many shares were found to
2291 have corruption, summed over all
2292 objects examined (after any
2297 """Return a dictionary with the same keys as
2298 IDirectoryNode.deep_stats()."""
2300 def get_corrupt_shares():
2301 """Return a set of (serverid, storage_index, sharenum) for all shares
2302 that were found to be corrupt before any repair was attempted. Both
2303 serverid and storage_index are binary.
2305 def get_remaining_corrupt_shares():
2306 """Return a set of (serverid, storage_index, sharenum) for all shares
2307 that were found to be corrupt after any repair was completed. Both
2308 serverid and storage_index are binary. These are shares that need
2309 manual inspection and probably deletion.
2311 def get_all_results():
2312 """Return a dictionary mapping pathname (a tuple of strings, ready to
2313 be slash-joined) to an ICheckAndRepairResults instance, one for each
2314 object that was checked."""
2316 def get_results_for_storage_index(storage_index):
2317 """Retrive the ICheckAndRepairResults instance for the given (binary)
2318 storage index. Raises KeyError if there are no results for that
2322 class IRepairable(Interface):
2323 def repair(check_results):
2324 """Attempt to repair the given object. Returns a Deferred that fires
2325 with a IRepairResults object.
2327 I must be called with an object that implements ICheckResults, as
2328 proof that you have actually discovered a problem with this file. I
2329 will use the data in the checker results to guide the repair process,
2330 such as which servers provided bad data and should therefore be
2331 avoided. The ICheckResults object is inside the
2332 ICheckAndRepairResults object, which is returned by the
2333 ICheckable.check() method::
2335 d = filenode.check(repair=False)
2336 def _got_results(check_and_repair_results):
2337 check_results = check_and_repair_results.get_pre_repair_results()
2338 return filenode.repair(check_results)
2339 d.addCallback(_got_results)
2343 class IRepairResults(Interface):
2344 """I contain the results of a repair operation."""
2345 def get_successful(self):
2346 """Returns a boolean: True if the repair made the file healthy, False
2347 if not. Repair failure generally indicates a file that has been
2348 damaged beyond repair."""
2351 class IClient(Interface):
2352 def upload(uploadable):
2353 """Upload some data into a CHK, get back the UploadResults for it.
2354 @param uploadable: something that implements IUploadable
2355 @return: a Deferred that fires with the UploadResults instance.
2356 To get the URI for this file, use results.uri .
2359 def create_mutable_file(contents=""):
2360 """Create a new mutable file (with initial) contents, get back the
2363 @param contents: (bytestring, callable, or None): this provides the
2364 initial contents of the mutable file. If 'contents' is a bytestring,
2365 it will be used as-is. If 'contents' is a callable, it will be
2366 invoked with the new MutableFileNode instance and is expected to
2367 return a bytestring with the initial contents of the file (the
2368 callable can use node.get_writekey() to decide how to encrypt the
2369 initial contents, e.g. for a brand new dirnode with initial
2370 children). contents=None is equivalent to an empty string. Using
2371 content_maker= is more efficient than creating a mutable file and
2372 setting its contents in two separate operations.
2374 @return: a Deferred that fires with an IMutableFileNode instance.
2377 def create_dirnode(initial_children={}):
2378 """Create a new unattached dirnode, possibly with initial children.
2380 @param initial_children: dict with keys that are unicode child names,
2381 and values that are (childnode, metadata) tuples.
2383 @return: a Deferred that fires with the new IDirectoryNode instance.
2386 def create_node_from_uri(uri, rouri):
2387 """Create a new IFilesystemNode instance from the uri, synchronously.
2388 @param uri: a string or IURI-providing instance, or None. This could
2389 be for a LiteralFileNode, a CHK file node, a mutable file
2390 node, or a directory node
2391 @param rouri: a string or IURI-providing instance, or None. If the
2392 main uri is None, I will use the rouri instead. If I
2393 recognize the format of the main uri, I will ignore the
2394 rouri (because it can be derived from the writecap).
2396 @return: an instance that provides IFilesystemNode (or more usefully
2397 one of its subclasses). File-specifying URIs will result in
2398 IFileNode-providing instances, like ImmutableFileNode,
2399 LiteralFileNode, or MutableFileNode. Directory-specifying
2400 URIs will result in IDirectoryNode-providing instances, like
2404 class INodeMaker(Interface):
2405 """The NodeMaker is used to create IFilesystemNode instances. It can
2406 accept a filecap/dircap string and return the node right away. It can
2407 also create new nodes (i.e. upload a file, or create a mutable file)
2408 asynchronously. Once you have one of these nodes, you can use other
2409 methods to determine whether it is a file or directory, and to download
2410 or modify its contents.
2412 The NodeMaker encapsulates all the authorities that these
2413 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2414 Tahoe process will typically have a single NodeMaker, but unit tests may
2415 create simplified/mocked forms for testing purposes.
2417 def create_from_cap(writecap, readcap=None, **kwargs):
2418 """I create an IFilesystemNode from the given writecap/readcap. I can
2419 only provide nodes for existing file/directory objects: use my other
2420 methods to create new objects. I return synchronously."""
2422 def create_mutable_file(contents=None, keysize=None):
2423 """I create a new mutable file, and return a Deferred which will fire
2424 with the IMutableFileNode instance when it is ready. If contents= is
2425 provided (a bytestring), it will be used as the initial contents of
2426 the new file, otherwise the file will contain zero bytes. keysize= is
2427 for use by unit tests, to create mutable files that are smaller than
2430 def create_new_mutable_directory(initial_children={}):
2431 """I create a new mutable directory, and return a Deferred which will
2432 fire with the IDirectoryNode instance when it is ready. If
2433 initial_children= is provided (a dict mapping unicode child name to
2434 (childnode, metadata_dict) tuples), the directory will be populated
2435 with those children, otherwise it will be empty."""
2437 class IClientStatus(Interface):
2438 def list_all_uploads():
2439 """Return a list of uploader objects, one for each upload which
2440 currently has an object available (tracked with weakrefs). This is
2441 intended for debugging purposes."""
2442 def list_active_uploads():
2443 """Return a list of active IUploadStatus objects."""
2444 def list_recent_uploads():
2445 """Return a list of IUploadStatus objects for the most recently
2448 def list_all_downloads():
2449 """Return a list of downloader objects, one for each download which
2450 currently has an object available (tracked with weakrefs). This is
2451 intended for debugging purposes."""
2452 def list_active_downloads():
2453 """Return a list of active IDownloadStatus objects."""
2454 def list_recent_downloads():
2455 """Return a list of IDownloadStatus objects for the most recently
2456 started downloads."""
2458 class IUploadStatus(Interface):
2460 """Return a timestamp (float with seconds since epoch) indicating
2461 when the operation was started."""
2462 def get_storage_index():
2463 """Return a string with the (binary) storage index in use on this
2464 upload. Returns None if the storage index has not yet been
2467 """Return an integer with the number of bytes that will eventually
2468 be uploaded for this file. Returns None if the size is not yet known.
2471 """Return True if this upload is using a Helper, False if not."""
2473 """Return a string describing the current state of the upload
2476 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2477 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2478 made towards hashing the file to determine a CHK encryption key: if
2479 non-convergent encryption is in use, this will be trivial, otherwise
2480 the whole file must be hashed. 'ciphertext' describes how much of the
2481 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2482 uploads. 'encode_and_push' describes how much of the encode-and-push
2483 process has finished: for helper uploads this is dependent upon the
2484 helper providing progress reports. It might be reasonable to add all
2485 three numbers and report the sum to the user."""
2487 """Return True if the upload is currently active, False if not."""
2489 """Return an instance of UploadResults (which contains timing and
2490 sharemap information). Might return None if the upload is not yet
2493 """Each upload status gets a unique number: this method returns that
2494 number. This provides a handle to this particular upload, so a web
2495 page can generate a suitable hyperlink."""
2497 class IDownloadStatus(Interface):
2499 """Return a timestamp (float with seconds since epoch) indicating
2500 when the operation was started."""
2501 def get_storage_index():
2502 """Return a string with the (binary) storage index in use on this
2503 download. This may be None if there is no storage index (i.e. LIT
2506 """Return an integer with the number of bytes that will eventually be
2507 retrieved for this file. Returns None if the size is not yet known.
2510 """Return True if this download is using a Helper, False if not."""
2512 """Return a string describing the current state of the download
2515 """Returns a float (from 0.0 to 1.0) describing the amount of the
2516 download that has completed. This value will remain at 0.0 until the
2517 first byte of plaintext is pushed to the download target."""
2519 """Return True if the download is currently active, False if not."""
2521 """Each download status gets a unique number: this method returns
2522 that number. This provides a handle to this particular download, so a
2523 web page can generate a suitable hyperlink."""
2525 class IServermapUpdaterStatus(Interface):
2527 class IPublishStatus(Interface):
2529 class IRetrieveStatus(Interface):
2532 class NotCapableError(Exception):
2533 """You have tried to write to a read-only node."""
2535 class BadWriteEnablerError(Exception):
2538 class RIControlClient(RemoteInterface):
2540 def wait_for_client_connections(num_clients=int):
2541 """Do not return until we have connections to at least NUM_CLIENTS
2545 def upload_from_file_to_uri(filename=str,
2546 convergence=ChoiceOf(None,
2547 StringConstraint(2**20))):
2548 """Upload a file to the grid. This accepts a filename (which must be
2549 absolute) that points to a file on the node's local disk. The node will
2550 read the contents of this file, upload it to the grid, then return the
2551 URI at which it was uploaded. If convergence is None then a random
2552 encryption key will be used, else the plaintext will be hashed, then
2553 that hash will be mixed together with the "convergence" string to form
2558 def download_from_uri_to_file(uri=URI, filename=str):
2559 """Download a file from the grid, placing it on the node's local disk
2560 at the given filename (which must be absolute[?]). Returns the
2561 absolute filename where the file was written."""
2566 def get_memory_usage():
2567 """Return a dict describes the amount of memory currently in use. The
2568 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2569 measuring memory consupmtion in bytes."""
2570 return DictOf(str, int)
2572 def speed_test(count=int, size=int, mutable=Any()):
2573 """Write 'count' tempfiles to disk, all of the given size. Measure
2574 how long (in seconds) it takes to upload them all to the servers.
2575 Then measure how long it takes to download all of them. If 'mutable'
2576 is 'create', time creation of mutable files. If 'mutable' is
2577 'upload', then time access to the same mutable file instead of
2580 Returns a tuple of (upload_time, download_time).
2582 return (float, float)
2584 def measure_peer_response_time():
2585 """Send a short message to each connected peer, and measure the time
2586 it takes for them to respond to it. This is a rough measure of the
2587 application-level round trip time.
2589 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2592 return DictOf(str, float)
2594 UploadResults = Any() #DictOf(str, str)
2596 class RIEncryptedUploadable(RemoteInterface):
2597 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2602 def get_all_encoding_parameters():
2603 return (int, int, int, long)
2605 def read_encrypted(offset=Offset, length=ReadSize):
2612 class RICHKUploadHelper(RemoteInterface):
2613 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2617 Return a dictionary of version information.
2619 return DictOf(str, Any())
2621 def upload(reader=RIEncryptedUploadable):
2622 return UploadResults
2625 class RIHelper(RemoteInterface):
2626 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2630 Return a dictionary of version information.
2632 return DictOf(str, Any())
2634 def upload_chk(si=StorageIndex):
2635 """See if a file with a given storage index needs uploading. The
2636 helper will ask the appropriate storage servers to see if the file
2637 has already been uploaded. If so, the helper will return a set of
2638 'upload results' that includes whatever hashes are needed to build
2639 the read-cap, and perhaps a truncated sharemap.
2641 If the file has not yet been uploaded (or if it was only partially
2642 uploaded), the helper will return an empty upload-results dictionary
2643 and also an RICHKUploadHelper object that will take care of the
2644 upload process. The client should call upload() on this object and
2645 pass it a reference to an RIEncryptedUploadable object that will
2646 provide ciphertext. When the upload is finished, the upload() method
2647 will finish and return the upload results.
2649 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2652 class RIStatsProvider(RemoteInterface):
2653 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2655 Provides access to statistics and monitoring information.
2660 returns a dictionary containing 'counters' and 'stats', each a
2661 dictionary with string counter/stat name keys, and numeric or None values.
2662 counters are monotonically increasing measures of work done, and
2663 stats are instantaneous measures (potentially time averaged
2666 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2668 class RIStatsGatherer(RemoteInterface):
2669 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2671 Provides a monitoring service for centralised collection of stats
2674 def provide(provider=RIStatsProvider, nickname=str):
2676 @param provider: a stats collector instance which should be polled
2677 periodically by the gatherer to collect stats.
2678 @param nickname: a name useful to identify the provided client
2683 class IStatsProducer(Interface):
2686 returns a dictionary, with str keys representing the names of stats
2687 to be monitored, and numeric values.
2690 class RIKeyGenerator(RemoteInterface):
2691 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2693 Provides a service offering to make RSA key pairs.
2696 def get_rsa_key_pair(key_size=int):
2698 @param key_size: the size of the signature key.
2699 @return: tuple(verifying_key, signing_key)
2701 return TupleOf(str, str)
2704 class FileTooLargeError(Exception):
2707 class IValidatedThingProxy(Interface):
2709 """ Acquire a thing and validate it. Return a deferred which is
2710 eventually fired with self if the thing is valid or errbacked if it
2711 can't be acquired or validated."""
2713 class InsufficientVersionError(Exception):
2714 def __init__(self, needed, got):
2715 self.needed = needed
2718 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2721 class EmptyPathnameComponentError(Exception):
2722 """The webapi disallows empty pathname components."""