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 share modifications
30 LeaseRenewSecret = Hash # used to protect lease renewal requests
31 LeaseCancelSecret = Hash # was used to protect lease cancellation requests
34 class RIBucketWriter(RemoteInterface):
35 """ Objects of this kind live on the server side. """
36 def write(offset=Offset, data=ShareData):
41 If the data that has been written is incomplete or inconsistent then
42 the server will throw the data away, else it will store it for future
48 """Abandon all the data that has been written.
53 class RIBucketReader(RemoteInterface):
54 def read(offset=Offset, length=ReadSize):
57 def advise_corrupt_share(reason=str):
58 """Clients who discover hash failures in shares that they have
59 downloaded from me will use this method to inform me about the
60 failures. I will record their concern so that my operator can
61 manually inspect the shares in question. I return None.
63 This is a wrapper around RIStorageServer.advise_corrupt_share()
64 that is tied to a specific share, and therefore does not need the
65 extra share-identifying arguments. Please see that method for full
70 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
71 # elements are (offset, length, operator, specimen)
72 # operator is one of "lt, le, eq, ne, ge, gt"
73 # nop always passes and is used to fetch data while writing.
74 # you should use length==len(specimen) for everything except nop
75 DataVector = ListOf(TupleOf(Offset, ShareData))
76 # (offset, data). This limits us to 30 writes of 1MiB each per call
77 TestAndWriteVectorsForShares = DictOf(int,
80 ChoiceOf(None, Offset), # new_length
82 ReadVector = ListOf(TupleOf(Offset, ReadSize))
83 ReadData = ListOf(ShareData)
84 # returns data[offset:offset+length] for each element of TestVector
87 class RIStorageServer(RemoteInterface):
88 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
92 Return a dictionary of version information.
94 return DictOf(str, Any())
96 def allocate_buckets(storage_index=StorageIndex,
97 renew_secret=LeaseRenewSecret,
98 cancel_secret=LeaseCancelSecret,
99 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
100 allocated_size=Offset, canary=Referenceable):
102 @param storage_index: the index of the bucket to be created or
104 @param sharenums: these are the share numbers (probably between 0 and
105 99) that the sender is proposing to store on this
107 @param renew_secret: This is the secret used to protect bucket refresh
108 This secret is generated by the client and
109 stored for later comparison by the server. Each
110 server is given a different secret.
111 @param cancel_secret: This no longer allows lease cancellation, but
112 must still be a unique value identifying the
113 lease. XXX stop relying on it to be unique.
114 @param canary: If the canary is lost before close(), the bucket is
116 @return: tuple of (alreadygot, allocated), where alreadygot is what we
117 already have and allocated is what we hereby agree to accept.
118 New leases are added for shares in both lists.
120 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
121 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
123 def add_lease(storage_index=StorageIndex,
124 renew_secret=LeaseRenewSecret,
125 cancel_secret=LeaseCancelSecret):
127 Add a new lease on the given bucket. If the renew_secret matches an
128 existing lease, that lease will be renewed instead. If there is no
129 bucket for the given storage_index, return silently. (note that in
130 tahoe-1.3.0 and earlier, IndexError was raised if there was no
133 return Any() # returns None now, but future versions might change
135 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
137 Renew the lease on a given bucket, resetting the timer to 31 days.
138 Some networks will use this, some will not. If there is no bucket for
139 the given storage_index, IndexError will be raised.
141 For mutable shares, if the given renew_secret does not match an
142 existing lease, IndexError will be raised with a note listing the
143 server-nodeids on the existing leases, so leases on migrated shares
144 can be renewed. For immutable shares, IndexError (without the note)
149 def get_buckets(storage_index=StorageIndex):
150 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
154 def slot_readv(storage_index=StorageIndex,
155 shares=ListOf(int), readv=ReadVector):
156 """Read a vector from the numbered shares associated with the given
157 storage index. An empty shares list means to return data from all
158 known shares. Returns a dictionary with one key per share."""
159 return DictOf(int, ReadData) # shnum -> results
161 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
162 secrets=TupleOf(WriteEnablerSecret,
165 tw_vectors=TestAndWriteVectorsForShares,
169 General-purpose test-read-and-set operation for mutable slots:
170 (1) For submitted shnums, compare the test vectors against extant
171 shares, or against an empty share for shnums that do not exist.
172 (2) Use the read vectors to extract "old data" from extant shares.
173 (3) If all tests in (1) passed, then apply the write vectors
174 (possibly creating new shares).
175 (4) Return whether the tests passed, and the "old data", which does
176 not include any modifications made by the writes.
178 The operation does not interleave with other operations on the same
181 This method is, um, large. The goal is to allow clients to update all
182 the shares associated with a mutable file in a single round trip.
184 @param storage_index: the index of the bucket to be created or
186 @param write_enabler: a secret that is stored along with the slot.
187 Writes are accepted from any caller who can
188 present the matching secret. A different secret
189 should be used for each slot*server pair.
190 @param renew_secret: This is the secret used to protect bucket refresh
191 This secret is generated by the client and
192 stored for later comparison by the server. Each
193 server is given a different secret.
194 @param cancel_secret: This no longer allows lease cancellation, but
195 must still be a unique value identifying the
196 lease. XXX stop relying on it to be unique.
198 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
199 cancel_secret). The first is required to perform any write. The
200 latter two are used when allocating new shares. To simply acquire a
201 new lease on existing shares, use an empty testv and an empty writev.
203 Each share can have a separate test vector (i.e. a list of
204 comparisons to perform). If all vectors for all shares pass, then all
205 writes for all shares are recorded. Each comparison is a 4-tuple of
206 (offset, length, operator, specimen), which effectively does a
207 bool( (read(offset, length)) OPERATOR specimen ) and only performs
208 the write if all these evaluate to True. Basic test-and-set uses 'eq'.
209 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
210 Write-if-same-or-newer uses 'le'.
212 Reads from the end of the container are truncated, and missing shares
213 behave like empty ones, so to assert that a share doesn't exist (for
214 use when creating a new share), use (0, 1, 'eq', '').
216 The write vector will be applied to the given share, expanding it if
217 necessary. A write vector applied to a share number that did not
218 exist previously will cause that share to be created. Write vectors
219 must not overlap (if they do, this will either cause an error or
220 apply them in an unspecified order). Duplicate write vectors, with
221 the same offset and data, are currently tolerated but are not
224 In Tahoe-LAFS v1.8.3 or later (except 1.9.0a1), if you send a write
225 vector whose offset is beyond the end of the current data, the space
226 between the end of the current data and the beginning of the write
227 vector will be filled with zero bytes. In earlier versions the
228 contents of this space was unspecified (and might end up containing
229 secrets). Storage servers with the new zero-filling behavior will
230 advertise a true value for the 'fills-holes-with-zero-bytes' key
231 (under 'http://allmydata.org/tahoe/protocols/storage/v1') in their
234 Each write vector is accompanied by a 'new_length' argument, which
235 can be used to truncate the data. If new_length is not None and it is
236 less than the current size of the data (after applying all write
237 vectors), then the data will be truncated to new_length. If
238 new_length==0, the share will be deleted.
240 In Tahoe-LAFS v1.8.2 and earlier, new_length could also be used to
241 enlarge the file by sending a number larger than the size of the data
242 after applying all write vectors. That behavior was not used, and as
243 of Tahoe-LAFS v1.8.3 it no longer works and the new_length is ignored
246 If a storage client knows that the server supports zero-filling, for
247 example from the 'fills-holes-with-zero-bytes' key in its version
248 information, it can extend the file efficiently by writing a single
249 zero byte just before the new end-of-file. Otherwise it must
250 explicitly write zeroes to all bytes between the old and new
251 end-of-file. In any case it should avoid sending new_length larger
252 than the size of the data after applying all write vectors.
254 The read vector is used to extract data from all known shares,
255 *before* any writes have been applied. The same read vector is used
256 for all shares. This captures the state that was tested by the test
257 vector, for extant shares.
259 This method returns two values: a boolean and a dict. The boolean is
260 True if the write vectors were applied, False if not. The dict is
261 keyed by share number, and each value contains a list of strings, one
262 for each element of the read vector.
264 If the write_enabler is wrong, this will raise BadWriteEnablerError.
265 To enable share migration (using update_write_enabler), the exception
266 will have the nodeid used for the old write enabler embedded in it,
267 in the following string::
269 The write enabler was recorded by nodeid '%s'.
271 Note that the nodeid here is encoded using the same base32 encoding
272 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
274 return TupleOf(bool, DictOf(int, ReadData))
276 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
277 shnum=int, reason=str):
278 """Clients who discover hash failures in shares that they have
279 downloaded from me will use this method to inform me about the
280 failures. I will record their concern so that my operator can
281 manually inspect the shares in question. I return None.
283 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
284 (binary) storage index string, and 'shnum' is the integer share
285 number. 'reason' is a human-readable explanation of the problem,
286 probably including some expected hash values and the computed ones
287 that did not match. Corruption advisories for mutable shares should
288 include a hash of the public key (the same value that appears in the
289 mutable-file verify-cap), since the current share format does not
294 class IStorageBucketWriter(Interface):
296 Objects of this kind live on the client side.
298 def put_block(segmentnum, data):
300 @param segmentnum=int
301 @param data=ShareData: For most segments, this data will be 'blocksize'
302 bytes in length. The last segment might be shorter.
303 @return: a Deferred that fires (with None) when the operation completes
306 def put_crypttext_hashes(hashes):
308 @param hashes=ListOf(Hash)
309 @return: a Deferred that fires (with None) when the operation completes
312 def put_block_hashes(blockhashes):
314 @param blockhashes=ListOf(Hash)
315 @return: a Deferred that fires (with None) when the operation completes
318 def put_share_hashes(sharehashes):
320 @param sharehashes=ListOf(TupleOf(int, Hash))
321 @return: a Deferred that fires (with None) when the operation completes
324 def put_uri_extension(data):
325 """This block of data contains integrity-checking information (hashes
326 of plaintext, crypttext, and shares), as well as encoding parameters
327 that are necessary to recover the data. This is a serialized dict
328 mapping strings to other strings. The hash of this data is kept in
329 the URI and verified before any of the data is used. All buckets for
330 a given file contain identical copies of this data.
332 The serialization format is specified with the following pseudocode:
333 for k in sorted(dict.keys()):
334 assert re.match(r'^[a-zA-Z_\-]+$', k)
335 write(k + ':' + netstring(dict[k]))
337 @param data=URIExtensionData
338 @return: a Deferred that fires (with None) when the operation completes
342 """Finish writing and close the bucket. The share is not finalized
343 until this method is called: if the uploading client disconnects
344 before calling close(), the partially-written share will be
347 @return: a Deferred that fires (with None) when the operation completes
351 class IStorageBucketReader(Interface):
352 def get_block_data(blocknum, blocksize, size):
353 """Most blocks will be the same size. The last block might be shorter
362 def get_crypttext_hashes():
364 @return: ListOf(Hash)
367 def get_block_hashes(at_least_these=()):
369 @param at_least_these=SetOf(int)
370 @return: ListOf(Hash)
373 def get_share_hashes():
375 @return: ListOf(TupleOf(int, Hash))
378 def get_uri_extension():
380 @return: URIExtensionData
384 class IStorageBroker(Interface):
385 def get_servers_for_psi(peer_selection_index):
387 @return: list of IServer instances
389 def get_connected_servers():
391 @return: frozenset of connected IServer instances
393 def get_known_servers():
395 @return: frozenset of IServer instances
397 def get_all_serverids():
399 @return: frozenset of serverid strings
401 def get_nickname_for_serverid(serverid):
403 @return: unicode nickname, or None
406 # methods moved from IntroducerClient, need review
407 def get_all_connections():
408 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
409 each active connection we've established to a remote service. This is
410 mostly useful for unit tests that need to wait until a certain number
411 of connections have been made."""
413 def get_all_connectors():
414 """Return a dict that maps from (nodeid, service_name) to a
415 RemoteServiceConnector instance for all services that we are actively
416 trying to connect to. Each RemoteServiceConnector has the following
419 service_name: the type of service provided, like 'storage'
420 announcement_time: when we first heard about this service
421 last_connect_time: when we last established a connection
422 last_loss_time: when we last lost a connection
424 version: the peer's version, from the most recent connection
425 oldest_supported: the peer's oldest supported version, same
427 rref: the RemoteReference, if connected, otherwise None
428 remote_host: the IAddress, if connected, otherwise None
430 This method is intended for monitoring interfaces, such as a web page
431 that describes connecting and connected peers.
434 def get_all_peerids():
435 """Return a frozenset of all peerids to whom we have a connection (to
436 one or more services) established. Mostly useful for unit tests."""
438 def get_all_connections_for(service_name):
439 """Return a frozenset of (nodeid, service_name, rref) tuples, one
440 for each active connection that provides the given SERVICE_NAME."""
442 def get_permuted_peers(service_name, key):
443 """Returns an ordered list of (peerid, rref) tuples, selecting from
444 the connections that provide SERVICE_NAME, using a hash-based
445 permutation keyed by KEY. This randomizes the service list in a
446 repeatable way, to distribute load over many peers.
450 class IDisplayableServer(Interface):
461 class IServer(IDisplayableServer):
462 """I live in the client, and represent a single server."""
463 def start_connecting(tub, trigger_cb):
467 """Once a server is connected, I return a RemoteReference.
468 Before a server is connected for the first time, I return None.
470 Note that the rref I return will start producing DeadReferenceErrors
471 once the connection is lost.
475 class IMutableSlotWriter(Interface):
477 The interface for a writer around a mutable slot on a remote server.
479 def set_checkstring(seqnum_or_checkstring, root_hash=None, salt=None):
481 Set the checkstring that I will pass to the remote server when
484 @param checkstring A packed checkstring to use.
486 Note that implementations can differ in which semantics they
487 wish to support for set_checkstring -- they can, for example,
488 build the checkstring themselves from its constituents, or
492 def get_checkstring():
494 Get the checkstring that I think currently exists on the remote
498 def put_block(data, segnum, salt):
500 Add a block and salt to the share.
503 def put_encprivkey(encprivkey):
505 Add the encrypted private key to the share.
508 def put_blockhashes(blockhashes):
510 @param blockhashes=list
511 Add the block hash tree to the share.
514 def put_sharehashes(sharehashes):
516 @param sharehashes=dict
517 Add the share hash chain to the share.
522 Return the part of the share that needs to be signed.
525 def put_signature(signature):
527 Add the signature to the share.
530 def put_verification_key(verification_key):
532 Add the verification key to the share.
535 def finish_publishing():
537 Do anything necessary to finish writing the share to a remote
538 server. I require that no further publishing needs to take place
539 after this method has been called.
543 class IURI(Interface):
544 def init_from_string(uri):
545 """Accept a string (as created by my to_string() method) and populate
546 this instance with its data. I am not normally called directly,
547 please use the module-level uri.from_string() function to convert
548 arbitrary URI strings into IURI-providing instances."""
551 """Return False if this URI be used to modify the data. Return True
552 if this URI cannot be used to modify the data."""
555 """Return True if the data can be modified by *somebody* (perhaps
556 someone who has a more powerful URI than this one)."""
558 # TODO: rename to get_read_cap()
560 """Return another IURI instance that represents a read-only form of
561 this one. If is_readonly() is True, this returns self."""
563 def get_verify_cap():
564 """Return an instance that provides IVerifierURI, which can be used
565 to check on the availability of the file or directory, without
566 providing enough capabilities to actually read or modify the
567 contents. This may return None if the file does not need checking or
568 verification (e.g. LIT URIs).
572 """Return a string of printable ASCII characters, suitable for
573 passing into init_from_string."""
576 class IVerifierURI(Interface, IURI):
577 def init_from_string(uri):
578 """Accept a string (as created by my to_string() method) and populate
579 this instance with its data. I am not normally called directly,
580 please use the module-level uri.from_string() function to convert
581 arbitrary URI strings into IURI-providing instances."""
584 """Return a string of printable ASCII characters, suitable for
585 passing into init_from_string."""
588 class IDirnodeURI(Interface):
589 """I am a URI that represents a dirnode."""
592 class IFileURI(Interface):
593 """I am a URI that represents a filenode."""
595 """Return the length (in bytes) of the file that I represent."""
598 class IImmutableFileURI(IFileURI):
601 class IMutableFileURI(Interface):
604 class IDirectoryURI(Interface):
607 class IReadonlyDirectoryURI(Interface):
611 class CapConstraintError(Exception):
612 """A constraint on a cap was violated."""
614 class MustBeDeepImmutableError(CapConstraintError):
615 """Mutable children cannot be added to an immutable directory.
616 Also, caps obtained from an immutable directory can trigger this error
617 if they are later found to refer to a mutable object and then used."""
619 class MustBeReadonlyError(CapConstraintError):
620 """Known write caps cannot be specified in a ro_uri field. Also,
621 caps obtained from a ro_uri field can trigger this error if they
622 are later found to be write caps and then used."""
624 class MustNotBeUnknownRWError(CapConstraintError):
625 """Cannot add an unknown child cap specified in a rw_uri field."""
628 class IReadable(Interface):
629 """I represent a readable object -- either an immutable file, or a
630 specific version of a mutable file.
634 """Return True if this reference provides mutable access to the given
635 file or directory (i.e. if you can modify it), or False if not. Note
636 that even if this reference is read-only, someone else may hold a
637 read-write reference to it.
639 For an IReadable returned by get_best_readable_version(), this will
640 always return True, but for instances of subinterfaces such as
641 IMutableFileVersion, it may return False."""
644 """Return True if this file or directory is mutable (by *somebody*,
645 not necessarily you), False if it is is immutable. Note that a file
646 might be mutable overall, but your reference to it might be
647 read-only. On the other hand, all references to an immutable file
648 will be read-only; there are no read-write references to an immutable
651 def get_storage_index():
652 """Return the storage index of the file."""
655 """Return the length (in bytes) of this readable object."""
657 def download_to_data():
658 """Download all of the file contents. I return a Deferred that fires
659 with the contents as a byte string."""
661 def read(consumer, offset=0, size=None):
662 """Download a portion (possibly all) of the file's contents, making
663 them available to the given IConsumer. Return a Deferred that fires
664 (with the consumer) when the consumer is unregistered (either because
665 the last byte has been given to it, or because the consumer threw an
666 exception during write(), possibly because it no longer wants to
667 receive data). The portion downloaded will start at 'offset' and
668 contain 'size' bytes (or the remainder of the file if size==None).
670 The consumer will be used in non-streaming mode: an IPullProducer
671 will be attached to it.
673 The consumer will not receive data right away: several network trips
674 must occur first. The order of events will be::
676 consumer.registerProducer(p, streaming)
677 (if streaming == False)::
678 consumer does p.resumeProducing()
680 consumer does p.resumeProducing()
681 consumer.write(data).. (repeat until all data is written)
682 consumer.unregisterProducer()
683 deferred.callback(consumer)
685 If a download error occurs, or an exception is raised by
686 consumer.registerProducer() or consumer.write(), I will call
687 consumer.unregisterProducer() and then deliver the exception via
688 deferred.errback(). To cancel the download, the consumer should call
689 p.stopProducing(), which will result in an exception being delivered
690 via deferred.errback().
692 See src/allmydata/util/consumer.py for an example of a simple
693 download-to-memory consumer.
697 class IWriteable(Interface):
699 I define methods that callers can use to update SDMF and MDMF
700 mutable files on a Tahoe-LAFS grid.
702 # XXX: For the moment, we have only this. It is possible that we
703 # want to move overwrite() and modify() in here too.
704 def update(data, offset):
706 I write the data from my data argument to the MDMF file,
707 starting at offset. I continue writing data until my data
708 argument is exhausted, appending data to the file as necessary.
710 # assert IMutableUploadable.providedBy(data)
711 # to append data: offset=node.get_size_of_best_version()
712 # do we want to support compacting MDMF?
713 # for an MDMF file, this can be done with O(data.get_size())
714 # memory. For an SDMF file, any modification takes
715 # O(node.get_size_of_best_version()).
718 class IMutableFileVersion(IReadable):
719 """I provide access to a particular version of a mutable file. The
720 access is read/write if I was obtained from a filenode derived from
721 a write cap, or read-only if the filenode was derived from a read cap.
724 def get_sequence_number():
725 """Return the sequence number of this version."""
728 """Return the IMutableFileServerMap instance that was used to create
733 """Return this filenode's writekey, or None if the node does not have
734 write-capability. This may be used to assist with data structures
735 that need to make certain data available only to writers, such as the
736 read-write child caps in dirnodes. The recommended process is to have
737 reader-visible data be submitted to the filenode in the clear (where
738 it will be encrypted by the filenode using the readkey), but encrypt
739 writer-visible data using this writekey.
742 def overwrite(new_contents):
743 """Replace the contents of the mutable file, provided that no other
744 node has published (or is attempting to publish, concurrently) a
745 newer version of the file than this one.
747 I will avoid modifying any share that is different than the version
748 given by get_sequence_number(). However, if another node is writing
749 to the file at the same time as me, I may manage to update some shares
750 while they update others. If I see any evidence of this, I will signal
751 UncoordinatedWriteError, and the file will be left in an inconsistent
752 state (possibly the version you provided, possibly the old version,
753 possibly somebody else's version, and possibly a mix of shares from
756 The recommended response to UncoordinatedWriteError is to either
757 return it to the caller (since they failed to coordinate their
758 writes), or to attempt some sort of recovery. It may be sufficient to
759 wait a random interval (with exponential backoff) and repeat your
760 operation. If I do not signal UncoordinatedWriteError, then I was
761 able to write the new version without incident.
763 I return a Deferred that fires (with a PublishStatus object) when the
764 update has completed.
767 def modify(modifier_cb):
768 """Modify the contents of the file, by downloading this version,
769 applying the modifier function (or bound method), then uploading
770 the new version. This will succeed as long as no other node
771 publishes a version between the download and the upload.
772 I return a Deferred that fires (with a PublishStatus object) when
773 the update is complete.
775 The modifier callable will be given three arguments: a string (with
776 the old contents), a 'first_time' boolean, and a servermap. As with
777 download_to_data(), the old contents will be from this version,
778 but the modifier can use the servermap to make other decisions
779 (such as refusing to apply the delta if there are multiple parallel
780 versions, or if there is evidence of a newer unrecoverable version).
781 'first_time' will be True the first time the modifier is called,
782 and False on any subsequent calls.
784 The callable should return a string with the new contents. The
785 callable must be prepared to be called multiple times, and must
786 examine the input string to see if the change that it wants to make
787 is already present in the old version. If it does not need to make
788 any changes, it can either return None, or return its input string.
790 If the modifier raises an exception, it will be returned in the
795 # The hierarchy looks like this:
802 class IFilesystemNode(Interface):
804 """Return the strongest 'cap instance' associated with this node.
805 (writecap for writeable-mutable files/directories, readcap for
806 immutable or readonly-mutable files/directories). To convert this
807 into a string, call .to_string() on the result."""
810 """Return a readonly cap instance for this node. For immutable or
811 readonly nodes, get_cap() and get_readcap() return the same thing."""
813 def get_repair_cap():
814 """Return an IURI instance that can be used to repair the file, or
815 None if this node cannot be repaired (either because it is not
816 distributed, like a LIT file, or because the node does not represent
817 sufficient authority to create a repair-cap, like a read-only RSA
818 mutable file node [which cannot create the correct write-enablers]).
821 def get_verify_cap():
822 """Return an IVerifierURI instance that represents the
823 'verifiy/refresh capability' for this node. The holder of this
824 capability will be able to renew the lease for this node, protecting
825 it from garbage-collection. They will also be able to ask a server if
826 it holds a share for the file or directory.
830 """Return the URI string corresponding to the strongest cap associated
831 with this node. If this node is read-only, the URI will only offer
832 read-only access. If this node is read-write, the URI will offer
835 If you have read-write access to a node and wish to share merely
836 read-only access with others, use get_readonly_uri().
840 """Return the URI string that can be used by others to get write
841 access to this node, if it is writeable. If this is a read-only node,
844 def get_readonly_uri():
845 """Return the URI string that can be used by others to get read-only
846 access to this node. The result is a read-only URI, regardless of
847 whether this node is read-only or read-write.
849 If you have merely read-only access to this node, get_readonly_uri()
850 will return the same thing as get_uri().
853 def get_storage_index():
854 """Return a string with the (binary) storage index in use on this
855 download. This may be None if there is no storage index (i.e. LIT
856 files and directories)."""
859 """Return True if this reference provides mutable access to the given
860 file or directory (i.e. if you can modify it), or False if not. Note
861 that even if this reference is read-only, someone else may hold a
862 read-write reference to it."""
865 """Return True if this file or directory is mutable (by *somebody*,
866 not necessarily you), False if it is is immutable. Note that a file
867 might be mutable overall, but your reference to it might be
868 read-only. On the other hand, all references to an immutable file
869 will be read-only; there are no read-write references to an immutable
874 """Return True if this is an unknown node."""
876 def is_allowed_in_immutable_directory():
877 """Return True if this node is allowed as a child of a deep-immutable
878 directory. This is true if either the node is of a known-immutable type,
879 or it is unknown and read-only.
883 """Raise any error associated with this node."""
885 # XXX: These may not be appropriate outside the context of an IReadable.
887 """Return the length (in bytes) of the data this node represents. For
888 directory nodes, I return the size of the backing store. I return
889 synchronously and do not consult the network, so for mutable objects,
890 I will return the most recently observed size for the object, or None
891 if I don't remember a size. Use get_current_size, which returns a
892 Deferred, if you want more up-to-date information."""
894 def get_current_size():
895 """I return a Deferred that fires with the length (in bytes) of the
896 data this node represents.
900 class IFileNode(IFilesystemNode):
901 """I am a node that represents a file: a sequence of bytes. I am not a
902 container, like IDirectoryNode."""
903 def get_best_readable_version():
904 """Return a Deferred that fires with an IReadable for the 'best'
905 available version of the file. The IReadable provides only read
906 access, even if this filenode was derived from a write cap.
908 For an immutable file, there is only one version. For a mutable
909 file, the 'best' version is the recoverable version with the
910 highest sequence number. If no uncoordinated writes have occurred,
911 and if enough shares are available, then this will be the most
912 recent version that has been uploaded. If no version is recoverable,
913 the Deferred will errback with an UnrecoverableFileError.
916 def download_best_version():
917 """Download the contents of the version that would be returned
918 by get_best_readable_version(). This is equivalent to calling
919 download_to_data() on the IReadable given by that method.
921 I return a Deferred that fires with a byte string when the file
922 has been fully downloaded. To support streaming download, use
923 the 'read' method of IReadable. If no version is recoverable,
924 the Deferred will errback with an UnrecoverableFileError.
927 def get_size_of_best_version():
928 """Find the size of the version that would be returned by
929 get_best_readable_version().
931 I return a Deferred that fires with an integer. If no version
932 is recoverable, the Deferred will errback with an
933 UnrecoverableFileError.
937 class IImmutableFileNode(IFileNode, IReadable):
938 """I am a node representing an immutable file. Immutable files have
942 class IMutableFileNode(IFileNode):
943 """I provide access to a 'mutable file', which retains its identity
944 regardless of what contents are put in it.
946 The consistency-vs-availability problem means that there might be
947 multiple versions of a file present in the grid, some of which might be
948 unrecoverable (i.e. have fewer than 'k' shares). These versions are
949 loosely ordered: each has a sequence number and a hash, and any version
950 with seqnum=N was uploaded by a node that has seen at least one version
953 The 'servermap' (an instance of IMutableFileServerMap) is used to
954 describe the versions that are known to be present in the grid, and which
955 servers are hosting their shares. It is used to represent the 'state of
956 the world', and is used for this purpose by my test-and-set operations.
957 Downloading the contents of the mutable file will also return a
958 servermap. Uploading a new version into the mutable file requires a
959 servermap as input, and the semantics of the replace operation is
960 'replace the file with my new version if it looks like nobody else has
961 changed the file since my previous download'. Because the file is
962 distributed, this is not a perfect test-and-set operation, but it will do
963 its best. If the replace process sees evidence of a simultaneous write,
964 it will signal an UncoordinatedWriteError, so that the caller can take
968 Most readers will want to use the 'best' current version of the file, and
969 should use my 'download_best_version()' method.
971 To unconditionally replace the file, callers should use overwrite(). This
972 is the mode that user-visible mutable files will probably use.
974 To apply some delta to the file, call modify() with a callable modifier
975 function that can apply the modification that you want to make. This is
976 the mode that dirnodes will use, since most directory modification
977 operations can be expressed in terms of deltas to the directory state.
980 Three methods are available for users who need to perform more complex
981 operations. The first is get_servermap(), which returns an up-to-date
982 servermap using a specified mode. The second is download_version(), which
983 downloads a specific version (not necessarily the 'best' one). The third
984 is 'upload', which accepts new contents and a servermap (which must have
985 been updated with MODE_WRITE). The upload method will attempt to apply
986 the new contents as long as no other node has modified the file since the
987 servermap was updated. This might be useful to a caller who wants to
988 merge multiple versions into a single new one.
990 Note that each time the servermap is updated, a specific 'mode' is used,
991 which determines how many peers are queried. To use a servermap for my
992 replace() method, that servermap must have been updated in MODE_WRITE.
993 These modes are defined in allmydata.mutable.common, and consist of
994 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
995 allmydata/mutable/servermap.py for details about the differences.
997 Mutable files are currently limited in size (about 3.5MB max) and can
998 only be retrieved and updated all-at-once, as a single big string. Future
999 versions of our mutable files will remove this restriction.
1001 def get_best_mutable_version():
1002 """Return a Deferred that fires with an IMutableFileVersion for
1003 the 'best' available version of the file. The best version is
1004 the recoverable version with the highest sequence number. If no
1005 uncoordinated writes have occurred, and if enough shares are
1006 available, then this will be the most recent version that has
1009 If no version is recoverable, the Deferred will errback with an
1010 UnrecoverableFileError.
1013 def overwrite(new_contents):
1014 """Unconditionally replace the contents of the mutable file with new
1015 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
1016 is only appropriate to use when the new contents of the file are
1017 completely unrelated to the old ones, and you do not care about other
1020 I return a Deferred that fires (with a PublishStatus object) when the
1021 update has completed.
1024 def modify(modifier_cb):
1025 """Modify the contents of the file, by downloading the current
1026 version, applying the modifier function (or bound method), then
1027 uploading the new version. I return a Deferred that fires (with a
1028 PublishStatus object) when the update is complete.
1030 The modifier callable will be given three arguments: a string (with
1031 the old contents), a 'first_time' boolean, and a servermap. As with
1032 download_best_version(), the old contents will be from the best
1033 recoverable version, but the modifier can use the servermap to make
1034 other decisions (such as refusing to apply the delta if there are
1035 multiple parallel versions, or if there is evidence of a newer
1036 unrecoverable version). 'first_time' will be True the first time the
1037 modifier is called, and False on any subsequent calls.
1039 The callable should return a string with the new contents. The
1040 callable must be prepared to be called multiple times, and must
1041 examine the input string to see if the change that it wants to make
1042 is already present in the old version. If it does not need to make
1043 any changes, it can either return None, or return its input string.
1045 If the modifier raises an exception, it will be returned in the
1049 def get_servermap(mode):
1050 """Return a Deferred that fires with an IMutableFileServerMap
1051 instance, updated using the given mode.
1054 def download_version(servermap, version):
1055 """Download a specific version of the file, using the servermap
1056 as a guide to where the shares are located.
1058 I return a Deferred that fires with the requested contents, or
1059 errbacks with UnrecoverableFileError. Note that a servermap that was
1060 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1061 all versions (those modes stop querying servers as soon as they can
1062 fulfil their goals), so you may want to use MODE_CHECK (which checks
1063 everything) to get increased visibility.
1066 def upload(new_contents, servermap):
1067 """Replace the contents of the file with new ones. This requires a
1068 servermap that was previously updated with MODE_WRITE.
1070 I attempt to provide test-and-set semantics, in that I will avoid
1071 modifying any share that is different than the version I saw in the
1072 servermap. However, if another node is writing to the file at the
1073 same time as me, I may manage to update some shares while they update
1074 others. If I see any evidence of this, I will signal
1075 UncoordinatedWriteError, and the file will be left in an inconsistent
1076 state (possibly the version you provided, possibly the old version,
1077 possibly somebody else's version, and possibly a mix of shares from
1080 The recommended response to UncoordinatedWriteError is to either
1081 return it to the caller (since they failed to coordinate their
1082 writes), or to attempt some sort of recovery. It may be sufficient to
1083 wait a random interval (with exponential backoff) and repeat your
1084 operation. If I do not signal UncoordinatedWriteError, then I was
1085 able to write the new version without incident.
1087 I return a Deferred that fires (with a PublishStatus object) when the
1088 publish has completed. I will update the servermap in-place with the
1089 location of all new shares.
1093 """Return this filenode's writekey, or None if the node does not have
1094 write-capability. This may be used to assist with data structures
1095 that need to make certain data available only to writers, such as the
1096 read-write child caps in dirnodes. The recommended process is to have
1097 reader-visible data be submitted to the filenode in the clear (where
1098 it will be encrypted by the filenode using the readkey), but encrypt
1099 writer-visible data using this writekey.
1103 """Returns the mutable file protocol version."""
1106 class NotEnoughSharesError(Exception):
1107 """Download was unable to get enough shares"""
1109 class NoSharesError(Exception):
1110 """Download was unable to get any shares at all."""
1112 class DownloadStopped(Exception):
1115 class UploadUnhappinessError(Exception):
1116 """Upload was unable to satisfy 'servers_of_happiness'"""
1118 class UnableToFetchCriticalDownloadDataError(Exception):
1119 """I was unable to fetch some piece of critical data that is supposed to
1120 be identically present in all shares."""
1122 class NoServersError(Exception):
1123 """Upload wasn't given any servers to work with, usually indicating a
1124 network or Introducer problem."""
1126 class ExistingChildError(Exception):
1127 """A directory node was asked to add or replace a child that already
1128 exists, and overwrite= was set to False."""
1130 class NoSuchChildError(Exception):
1131 """A directory node was asked to fetch a child that does not exist."""
1133 # avoid UnicodeEncodeErrors when converting to str
1134 return self.__repr__()
1136 class ChildOfWrongTypeError(Exception):
1137 """An operation was attempted on a child of the wrong type (file or directory)."""
1140 class IDirectoryNode(IFilesystemNode):
1141 """I represent a filesystem node that is a container, with a
1142 name-to-child mapping, holding the tahoe equivalent of a directory. All
1143 child names are unicode strings, and all children are some sort of
1144 IFilesystemNode (a file, subdirectory, or unknown node).
1149 The dirnode ('1') URI returned by this method can be used in
1150 set_uri() on a different directory ('2') to 'mount' a reference to
1151 this directory ('1') under the other ('2'). This URI is just a
1152 string, so it can be passed around through email or other out-of-band
1156 def get_readonly_uri():
1158 The dirnode ('1') URI returned by this method can be used in
1159 set_uri() on a different directory ('2') to 'mount' a reference to
1160 this directory ('1') under the other ('2'). This URI is just a
1161 string, so it can be passed around through email or other out-of-band
1166 """I return a Deferred that fires with a dictionary mapping child
1167 name (a unicode string) to (node, metadata_dict) tuples, in which
1168 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1171 def has_child(name):
1172 """I return a Deferred that fires with a boolean, True if there
1173 exists a child of the given name, False if not. The child name must
1174 be a unicode string."""
1177 """I return a Deferred that fires with a specific named child node,
1178 which is an IFilesystemNode. The child name must be a unicode string.
1179 I raise NoSuchChildError if I do not have a child by that name."""
1181 def get_metadata_for(name):
1182 """I return a Deferred that fires with the metadata dictionary for
1183 a specific named child node. The child name must be a unicode string.
1184 This metadata is stored in the *edge*, not in the child, so it is
1185 attached to the parent dirnode rather than the child node.
1186 I raise NoSuchChildError if I do not have a child by that name."""
1188 def set_metadata_for(name, metadata):
1189 """I replace any existing metadata for the named child with the new
1190 metadata. The child name must be a unicode string. This metadata is
1191 stored in the *edge*, not in the child, so it is attached to the
1192 parent dirnode rather than the child node. I return a Deferred
1193 (that fires with this dirnode) when the operation is complete.
1194 I raise NoSuchChildError if I do not have a child by that name."""
1196 def get_child_at_path(path):
1197 """Transform a child path into an IFilesystemNode.
1199 I perform a recursive series of 'get' operations to find the named
1200 descendant node. I return a Deferred that fires with the node, or
1201 errbacks with NoSuchChildError if the node could not be found.
1203 The path can be either a single string (slash-separated) or a list of
1204 path-name elements. All elements must be unicode strings.
1207 def get_child_and_metadata_at_path(path):
1208 """Transform a child path into an IFilesystemNode and metadata.
1210 I am like get_child_at_path(), but my Deferred fires with a tuple of
1211 (node, metadata). The metadata comes from the last edge. If the path
1212 is empty, the metadata will be an empty dictionary.
1215 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1216 """I add a child (by writecap+readcap) at the specific name. I return
1217 a Deferred that fires when the operation finishes. If overwrite= is
1218 True, I will replace any existing child of the same name, otherwise
1219 an existing child will cause me to return ExistingChildError. The
1220 child name must be a unicode string.
1222 The child caps could be for a file, or for a directory. If you have
1223 both the writecap and readcap, you should provide both arguments.
1224 If you have only one cap and don't know whether it is read-only,
1225 provide it as the writecap argument and leave the readcap as None.
1226 If you have only one cap that is known to be read-only, provide it
1227 as the readcap argument and leave the writecap as None.
1228 The filecaps are typically obtained from an IFilesystemNode with
1229 get_uri() and get_readonly_uri().
1231 If metadata= is provided, I will use it as the metadata for the named
1232 edge. This will replace any existing metadata. If metadata= is left
1233 as the default value of None, I will set ['mtime'] to the current
1234 time, and I will set ['ctime'] to the current time if there was not
1235 already a child by this name present. This roughly matches the
1236 ctime/mtime semantics of traditional filesystems. See the
1237 "About the metadata" section of webapi.txt for futher information.
1239 If this directory node is read-only, the Deferred will errback with a
1240 NotWriteableError."""
1242 def set_children(entries, overwrite=True):
1243 """Add multiple children (by writecap+readcap) to a directory node.
1244 Takes a dictionary, with childname as keys and (writecap, readcap)
1245 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1246 a Deferred that fires (with this dirnode) when the operation
1247 finishes. This is equivalent to calling set_uri() multiple times, but
1248 is much more efficient. All child names must be unicode strings.
1251 def set_node(name, child, metadata=None, overwrite=True):
1252 """I add a child at the specific name. I return a Deferred that fires
1253 when the operation finishes. This Deferred will fire with the child
1254 node that was just added. I will replace any existing child of the
1255 same name. The child name must be a unicode string. The 'child'
1256 instance must be an instance providing IFilesystemNode.
1258 If metadata= is provided, I will use it as the metadata for the named
1259 edge. This will replace any existing metadata. If metadata= is left
1260 as the default value of None, I will set ['mtime'] to the current
1261 time, and I will set ['ctime'] to the current time if there was not
1262 already a child by this name present. This roughly matches the
1263 ctime/mtime semantics of traditional filesystems. See the
1264 "About the metadata" section of webapi.txt for futher information.
1266 If this directory node is read-only, the Deferred will errback with a
1267 NotWriteableError."""
1269 def set_nodes(entries, overwrite=True):
1270 """Add multiple children to a directory node. Takes a dict mapping
1271 unicode childname to (child_node, metdata) tuples. If metdata=None,
1272 the original metadata is left unmodified. Returns a Deferred that
1273 fires (with this dirnode) when the operation finishes. This is
1274 equivalent to calling set_node() multiple times, but is much more
1277 def add_file(name, uploadable, metadata=None, overwrite=True):
1278 """I upload a file (using the given IUploadable), then attach the
1279 resulting ImmutableFileNode to the directory at the given name. I set
1280 metadata the same way as set_uri and set_node. The child name must be
1283 I return a Deferred that fires (with the IFileNode of the uploaded
1284 file) when the operation completes."""
1286 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1287 """I remove the child at the specific name. I return a Deferred that
1288 fires when the operation finishes. The child name must be a unicode
1289 string. If must_exist is True and I do not have a child by that name,
1290 I raise NoSuchChildError. If must_be_directory is True and the child
1291 is a file, or if must_be_file is True and the child is a directory,
1292 I raise ChildOfWrongTypeError."""
1294 def create_subdirectory(name, initial_children={}, overwrite=True,
1295 mutable=True, mutable_version=None, metadata=None):
1296 """I create and attach a directory at the given name. The new
1297 directory can be empty, or it can be populated with children
1298 according to 'initial_children', which takes a dictionary in the same
1299 format as set_nodes (i.e. mapping unicode child name to (childnode,
1300 metadata) tuples). The child name must be a unicode string. I return
1301 a Deferred that fires (with the new directory node) when the
1302 operation finishes."""
1304 def move_child_to(current_child_name, new_parent, new_child_name=None,
1306 """I take one of my children and move them to a new parent. The child
1307 is referenced by name. On the new parent, the child will live under
1308 'new_child_name', which defaults to 'current_child_name'. TODO: what
1309 should we do about metadata? I return a Deferred that fires when the
1310 operation finishes. The child name must be a unicode string. I raise
1311 NoSuchChildError if I do not have a child by that name."""
1313 def build_manifest():
1314 """I generate a table of everything reachable from this directory.
1315 I also compute deep-stats as described below.
1317 I return a Monitor. The Monitor's results will be a dictionary with
1320 res['manifest']: a list of (path, cap) tuples for all nodes
1321 (directories and files) reachable from this one.
1322 'path' will be a tuple of unicode strings. The
1323 origin dirnode will be represented by an empty path
1325 res['verifycaps']: a list of (printable) verifycap strings, one for
1326 each reachable non-LIT node. This is a set:
1327 it will contain no duplicates.
1328 res['storage-index']: a list of (base32) storage index strings,
1329 one for each reachable non-LIT node. This is
1330 a set: it will contain no duplicates.
1331 res['stats']: a dictionary, the same that is generated by
1332 start_deep_stats() below.
1334 The Monitor will also have an .origin_si attribute with the (binary)
1335 storage index of the starting point.
1338 def start_deep_stats():
1339 """Return a Monitor, examining all nodes (directories and files)
1340 reachable from this one. The Monitor's results will be a dictionary
1341 with the following keys::
1343 count-immutable-files: count of how many CHK files are in the set
1344 count-mutable-files: same, for mutable files (does not include
1346 count-literal-files: same, for LIT files
1347 count-files: sum of the above three
1349 count-directories: count of directories
1351 size-immutable-files: total bytes for all CHK files in the set
1352 size-mutable-files (TODO): same, for current version of all mutable
1353 files, does not include directories
1354 size-literal-files: same, for LIT files
1355 size-directories: size of mutable files used by directories
1357 largest-directory: number of bytes in the largest directory
1358 largest-directory-children: number of children in the largest
1360 largest-immutable-file: number of bytes in the largest CHK file
1362 size-mutable-files is not yet implemented, because it would involve
1363 even more queries than deep_stats does.
1365 The Monitor will also have an .origin_si attribute with the (binary)
1366 storage index of the starting point.
1368 This operation will visit every directory node underneath this one,
1369 and can take a long time to run. On a typical workstation with good
1370 bandwidth, this can examine roughly 15 directories per second (and
1371 takes several minutes of 100% CPU for ~1700 directories).
1375 class ICodecEncoder(Interface):
1376 def set_params(data_size, required_shares, max_shares):
1377 """Set up the parameters of this encoder.
1379 This prepares the encoder to perform an operation that converts a
1380 single block of data into a number of shares, such that a future
1381 ICodecDecoder can use a subset of these shares to recover the
1382 original data. This operation is invoked by calling encode(). Once
1383 the encoding parameters are set up, the encode operation can be
1384 invoked multiple times.
1386 set_params() prepares the encoder to accept blocks of input data that
1387 are exactly 'data_size' bytes in length. The encoder will be prepared
1388 to produce 'max_shares' shares for each encode() operation (although
1389 see the 'desired_share_ids' to use less CPU). The encoding math will
1390 be chosen such that the decoder can get by with as few as
1391 'required_shares' of these shares and still reproduce the original
1392 data. For example, set_params(1000, 5, 5) offers no redundancy at
1393 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1395 Numerical Restrictions: 'data_size' is required to be an integral
1396 multiple of 'required_shares'. In general, the caller should choose
1397 required_shares and max_shares based upon their reliability
1398 requirements and the number of peers available (the total storage
1399 space used is roughly equal to max_shares*data_size/required_shares),
1400 then choose data_size to achieve the memory footprint desired (larger
1401 data_size means more efficient operation, smaller data_size means
1402 smaller memory footprint).
1404 In addition, 'max_shares' must be equal to or greater than
1405 'required_shares'. Of course, setting them to be equal causes
1406 encode() to degenerate into a particularly slow form of the 'split'
1409 See encode() for more details about how these parameters are used.
1411 set_params() must be called before any other ICodecEncoder methods
1416 """Return the 3-tuple of data_size, required_shares, max_shares"""
1418 def get_encoder_type():
1419 """Return a short string that describes the type of this encoder.
1421 There is required to be a global table of encoder classes. This method
1422 returns an index into this table; the value at this index is an
1423 encoder class, and this encoder is an instance of that class.
1426 def get_block_size():
1427 """Return the length of the shares that encode() will produce.
1430 def encode_proposal(data, desired_share_ids=None):
1431 """Encode some data.
1433 'data' must be a string (or other buffer object), and len(data) must
1434 be equal to the 'data_size' value passed earlier to set_params().
1436 This will return a Deferred that will fire with two lists. The first
1437 is a list of shares, each of which is a string (or other buffer
1438 object) such that len(share) is the same as what get_share_size()
1439 returned earlier. The second is a list of shareids, in which each is
1440 an integer. The lengths of the two lists will always be equal to each
1441 other. The user should take care to keep each share closely
1442 associated with its shareid, as one is useless without the other.
1444 The length of this output list will normally be the same as the value
1445 provided to the 'max_shares' parameter of set_params(). This may be
1446 different if 'desired_share_ids' is provided.
1448 'desired_share_ids', if provided, is required to be a sequence of
1449 ints, each of which is required to be >= 0 and < max_shares. If not
1450 provided, encode() will produce 'max_shares' shares, as if
1451 'desired_share_ids' were set to range(max_shares). You might use this
1452 if you initially thought you were going to use 10 peers, started
1453 encoding, and then two of the peers dropped out: you could use
1454 desired_share_ids= to skip the work (both memory and CPU) of
1455 producing shares for the peers that are no longer available.
1459 def encode(inshares, desired_share_ids=None):
1460 """Encode some data. This may be called multiple times. Each call is
1463 inshares is a sequence of length required_shares, containing buffers
1464 (i.e. strings), where each buffer contains the next contiguous
1465 non-overlapping segment of the input data. Each buffer is required to
1466 be the same length, and the sum of the lengths of the buffers is
1467 required to be exactly the data_size promised by set_params(). (This
1468 implies that the data has to be padded before being passed to
1469 encode(), unless of course it already happens to be an even multiple
1470 of required_shares in length.)
1472 Note: the requirement to break up your data into
1473 'required_shares' chunks of exactly the right length before
1474 calling encode() is surprising from point of view of a user
1475 who doesn't know how FEC works. It feels like an
1476 implementation detail that has leaked outside the abstraction
1477 barrier. Is there a use case in which the data to be encoded
1478 might already be available in pre-segmented chunks, such that
1479 it is faster or less work to make encode() take a list rather
1480 than splitting a single string?
1482 Yes, there is: suppose you are uploading a file with K=64,
1483 N=128, segsize=262,144. Then each in-share will be of size
1484 4096. If you use this .encode() API then your code could first
1485 read each successive 4096-byte chunk from the file and store
1486 each one in a Python string and store each such Python string
1487 in a Python list. Then you could call .encode(), passing that
1488 list as "inshares". The encoder would generate the other 64
1489 "secondary shares" and return to you a new list containing
1490 references to the same 64 Python strings that you passed in
1491 (as the primary shares) plus references to the new 64 Python
1494 (You could even imagine that your code could use readv() so
1495 that the operating system can arrange to get all of those
1496 bytes copied from the file into the Python list of Python
1497 strings as efficiently as possible instead of having a loop
1498 written in C or in Python to copy the next part of the file
1499 into the next string.)
1501 On the other hand if you instead use the .encode_proposal()
1502 API (above), then your code can first read in all of the
1503 262,144 bytes of the segment from the file into a Python
1504 string, then call .encode_proposal() passing the segment data
1505 as the "data" argument. The encoder would basically first
1506 split the "data" argument into a list of 64 in-shares of 4096
1507 byte each, and then do the same thing that .encode() does. So
1508 this would result in a little bit more copying of data and a
1509 little bit higher of a "maximum memory usage" during the
1510 process, although it might or might not make a practical
1511 difference for our current use cases.
1513 Note that "inshares" is a strange name for the parameter if
1514 you think of the parameter as being just for feeding in data
1515 to the codec. It makes more sense if you think of the result
1516 of this encoding as being the set of shares from inshares plus
1517 an extra set of "secondary shares" (or "check shares"). It is
1518 a surprising name! If the API is going to be surprising then
1519 the name should be surprising. If we switch to
1520 encode_proposal() above then we should also switch to an
1523 'desired_share_ids', if provided, is required to be a sequence of
1524 ints, each of which is required to be >= 0 and < max_shares. If not
1525 provided, encode() will produce 'max_shares' shares, as if
1526 'desired_share_ids' were set to range(max_shares). You might use this
1527 if you initially thought you were going to use 10 peers, started
1528 encoding, and then two of the peers dropped out: you could use
1529 desired_share_ids= to skip the work (both memory and CPU) of
1530 producing shares for the peers that are no longer available.
1532 For each call, encode() will return a Deferred that fires with two
1533 lists, one containing shares and the other containing the shareids.
1534 The get_share_size() method can be used to determine the length of
1535 the share strings returned by encode(). Each shareid is a small
1536 integer, exactly as passed into 'desired_share_ids' (or
1537 range(max_shares), if desired_share_ids was not provided).
1539 The shares and their corresponding shareids are required to be kept
1540 together during storage and retrieval. Specifically, the share data is
1541 useless by itself: the decoder needs to be told which share is which
1542 by providing it with both the shareid and the actual share data.
1544 This function will allocate an amount of memory roughly equal to::
1546 (max_shares - required_shares) * get_share_size()
1548 When combined with the memory that the caller must allocate to
1549 provide the input data, this leads to a memory footprint roughly
1550 equal to the size of the resulting encoded shares (i.e. the expansion
1551 factor times the size of the input segment).
1556 # returning a list of (shareidN,shareN) tuples instead of a pair of
1557 # lists (shareids..,shares..). Brian thought the tuples would
1558 # encourage users to keep the share and shareid together throughout
1559 # later processing, Zooko pointed out that the code to iterate
1560 # through two lists is not really more complicated than using a list
1561 # of tuples and there's also a performance improvement
1563 # having 'data_size' not required to be an integral multiple of
1564 # 'required_shares'. Doing this would require encode() to perform
1565 # padding internally, and we'd prefer to have any padding be done
1566 # explicitly by the caller. Yes, it is an abstraction leak, but
1567 # hopefully not an onerous one.
1570 class ICodecDecoder(Interface):
1571 def set_params(data_size, required_shares, max_shares):
1572 """Set the params. They have to be exactly the same ones that were
1573 used for encoding."""
1575 def get_needed_shares():
1576 """Return the number of shares needed to reconstruct the data.
1577 set_params() is required to be called before this."""
1579 def decode(some_shares, their_shareids):
1580 """Decode a partial list of shares into data.
1582 'some_shares' is required to be a sequence of buffers of sharedata, a
1583 subset of the shares returned by ICodecEncode.encode(). Each share is
1584 required to be of the same length. The i'th element of their_shareids
1585 is required to be the shareid of the i'th buffer in some_shares.
1587 This returns a Deferred that fires with a sequence of buffers. This
1588 sequence will contain all of the segments of the original data, in
1589 order. The sum of the lengths of all of the buffers will be the
1590 'data_size' value passed into the original ICodecEncode.set_params()
1591 call. To get back the single original input block of data, use
1592 ''.join(output_buffers), or you may wish to simply write them in
1593 order to an output file.
1595 Note that some of the elements in the result sequence may be
1596 references to the elements of the some_shares input sequence. In
1597 particular, this means that if those share objects are mutable (e.g.
1598 arrays) and if they are changed, then both the input (the
1599 'some_shares' parameter) and the output (the value given when the
1600 deferred is triggered) will change.
1602 The length of 'some_shares' is required to be exactly the value of
1603 'required_shares' passed into the original ICodecEncode.set_params()
1608 class IEncoder(Interface):
1609 """I take an object that provides IEncryptedUploadable, which provides
1610 encrypted data, and a list of shareholders. I then encode, hash, and
1611 deliver shares to those shareholders. I will compute all the necessary
1612 Merkle hash trees that are necessary to validate the crypttext that
1613 eventually comes back from the shareholders. I provide the URI Extension
1614 Block Hash, and the encoding parameters, both of which must be included
1617 I do not choose shareholders, that is left to the IUploader. I must be
1618 given a dict of RemoteReferences to storage buckets that are ready and
1619 willing to receive data.
1623 """Specify the number of bytes that will be encoded. This must be
1624 peformed before get_serialized_params() can be called.
1627 def set_encrypted_uploadable(u):
1628 """Provide a source of encrypted upload data. 'u' must implement
1629 IEncryptedUploadable.
1631 When this is called, the IEncryptedUploadable will be queried for its
1632 length and the storage_index that should be used.
1634 This returns a Deferred that fires with this Encoder instance.
1636 This must be performed before start() can be called.
1639 def get_param(name):
1640 """Return an encoding parameter, by name.
1642 'storage_index': return a string with the (16-byte truncated SHA-256
1643 hash) storage index to which these shares should be
1646 'share_counts': return a tuple describing how many shares are used:
1647 (needed_shares, servers_of_happiness, total_shares)
1649 'num_segments': return an int with the number of segments that
1652 'segment_size': return an int with the size of each segment.
1654 'block_size': return the size of the individual blocks that will
1655 be delivered to a shareholder's put_block() method. By
1656 knowing this, the shareholder will be able to keep all
1657 blocks in a single file and still provide random access
1658 when reading them. # TODO: can we avoid exposing this?
1660 'share_size': an int with the size of the data that will be stored
1661 on each shareholder. This is aggregate amount of data
1662 that will be sent to the shareholder, summed over all
1663 the put_block() calls I will ever make. It is useful to
1664 determine this size before asking potential
1665 shareholders whether they will grant a lease or not,
1666 since their answers will depend upon how much space we
1667 need. TODO: this might also include some amount of
1668 overhead, like the size of all the hashes. We need to
1669 decide whether this is useful or not.
1671 'serialized_params': a string with a concise description of the
1672 codec name and its parameters. This may be passed
1673 into the IUploadable to let it make sure that
1674 the same file encoded with different parameters
1675 will result in different storage indexes.
1677 Once this is called, set_size() and set_params() may not be called.
1680 def set_shareholders(shareholders, servermap):
1681 """Tell the encoder where to put the encoded shares. 'shareholders'
1682 must be a dictionary that maps share number (an integer ranging from
1683 0 to n-1) to an instance that provides IStorageBucketWriter.
1684 'servermap' is a dictionary that maps share number (as defined above)
1685 to a set of peerids. This must be performed before start() can be
1689 """Begin the encode/upload process. This involves reading encrypted
1690 data from the IEncryptedUploadable, encoding it, uploading the shares
1691 to the shareholders, then sending the hash trees.
1693 set_encrypted_uploadable() and set_shareholders() must be called
1694 before this can be invoked.
1696 This returns a Deferred that fires with a verify cap when the upload
1697 process is complete. The verifycap, plus the encryption key, is
1698 sufficient to construct the read cap.
1702 class IDecoder(Interface):
1703 """I take a list of shareholders and some setup information, then
1704 download, validate, decode, and decrypt data from them, writing the
1705 results to an output file.
1707 I do not locate the shareholders, that is left to the IDownloader. I must
1708 be given a dict of RemoteReferences to storage buckets that are ready to
1713 """I take a file-like object (providing write and close) to which all
1714 the plaintext data will be written.
1716 TODO: producer/consumer . Maybe write() should return a Deferred that
1717 indicates when it will accept more data? But probably having the
1718 IDecoder be a producer is easier to glue to IConsumer pieces.
1721 def set_shareholders(shareholders):
1722 """I take a dictionary that maps share identifiers (small integers)
1723 to RemoteReferences that provide RIBucketReader. This must be called
1727 """I start the download. This process involves retrieving data and
1728 hash chains from the shareholders, using the hashes to validate the
1729 data, decoding the shares into segments, decrypting the segments,
1730 then writing the resulting plaintext to the output file.
1732 I return a Deferred that will fire (with self) when the download is
1737 class IDownloadTarget(Interface):
1738 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1739 # will register itself as a producer. This allows the target to invoke
1740 # downloader.pauseProducing, resumeProducing, and stopProducing.
1742 """Called before any calls to write() or close(). If an error
1743 occurs before any data is available, fail() may be called without
1744 a previous call to open().
1746 'size' is the length of the file being downloaded, in bytes."""
1749 """Output some data to the target."""
1752 """Inform the target that there is no more data to be written."""
1755 """fail() is called to indicate that the download has failed. 'why'
1756 is a Failure object indicating what went wrong. No further methods
1757 will be invoked on the IDownloadTarget after fail()."""
1759 def register_canceller(cb):
1760 """The CiphertextDownloader uses this to register a no-argument function
1761 that the target can call to cancel the download. Once this canceller
1762 is invoked, no further calls to write() or close() will be made."""
1765 """When the CiphertextDownloader is done, this finish() function will be
1766 called. Whatever it returns will be returned to the invoker of
1767 Downloader.download.
1771 class IDownloader(Interface):
1772 def download(uri, target):
1773 """Perform a CHK download, sending the data to the given target.
1774 'target' must provide IDownloadTarget.
1776 Returns a Deferred that fires (with the results of target.finish)
1777 when the download is finished, or errbacks if something went wrong."""
1780 class IEncryptedUploadable(Interface):
1781 def set_upload_status(upload_status):
1782 """Provide an IUploadStatus object that should be filled with status
1783 information. The IEncryptedUploadable is responsible for setting
1784 key-determination progress ('chk'), size, storage_index, and
1785 ciphertext-fetch progress. It may delegate some of this
1786 responsibility to others, in particular to the IUploadable."""
1789 """This behaves just like IUploadable.get_size()."""
1791 def get_all_encoding_parameters():
1792 """Return a Deferred that fires with a tuple of
1793 (k,happy,n,segment_size). The segment_size will be used as-is, and
1794 must match the following constraints: it must be a multiple of k, and
1795 it shouldn't be unreasonably larger than the file size (if
1796 segment_size is larger than filesize, the difference must be stored
1799 This usually passes through to the IUploadable method of the same
1802 The encoder strictly obeys the values returned by this method. To
1803 make an upload use non-default encoding parameters, you must arrange
1804 to control the values that this method returns.
1807 def get_storage_index():
1808 """Return a Deferred that fires with a 16-byte storage index.
1811 def read_encrypted(length, hash_only):
1812 """This behaves just like IUploadable.read(), but returns crypttext
1813 instead of plaintext. If hash_only is True, then this discards the
1814 data (and returns an empty list); this improves efficiency when
1815 resuming an interrupted upload (where we need to compute the
1816 plaintext hashes, but don't need the redundant encrypted data)."""
1819 """Just like IUploadable.close()."""
1822 class IUploadable(Interface):
1823 def set_upload_status(upload_status):
1824 """Provide an IUploadStatus object that should be filled with status
1825 information. The IUploadable is responsible for setting
1826 key-determination progress ('chk')."""
1828 def set_default_encoding_parameters(params):
1829 """Set the default encoding parameters, which must be a dict mapping
1830 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1831 'max_segment_size'. These might have an influence on the final
1832 encoding parameters returned by get_all_encoding_parameters(), if the
1833 Uploadable doesn't have more specific preferences.
1835 This call is optional: if it is not used, the Uploadable will use
1836 some built-in defaults. If used, this method must be called before
1837 any other IUploadable methods to have any effect.
1841 """Return a Deferred that will fire with the length of the data to be
1842 uploaded, in bytes. This will be called before the data is actually
1843 used, to compute encoding parameters.
1846 def get_all_encoding_parameters():
1847 """Return a Deferred that fires with a tuple of
1848 (k,happy,n,segment_size). The segment_size will be used as-is, and
1849 must match the following constraints: it must be a multiple of k, and
1850 it shouldn't be unreasonably larger than the file size (if
1851 segment_size is larger than filesize, the difference must be stored
1854 The relative values of k and n allow some IUploadables to request
1855 better redundancy than others (in exchange for consuming more space
1858 Larger values of segment_size reduce hash overhead, while smaller
1859 values reduce memory footprint and cause data to be delivered in
1860 smaller pieces (which may provide a smoother and more predictable
1861 download experience).
1863 The encoder strictly obeys the values returned by this method. To
1864 make an upload use non-default encoding parameters, you must arrange
1865 to control the values that this method returns. One way to influence
1866 them may be to call set_encoding_parameters() before calling
1867 get_all_encoding_parameters().
1870 def get_encryption_key():
1871 """Return a Deferred that fires with a 16-byte AES key. This key will
1872 be used to encrypt the data. The key will also be hashed to derive
1875 Uploadables that want to achieve convergence should hash their file
1876 contents and the serialized_encoding_parameters to form the key
1877 (which of course requires a full pass over the data). Uploadables can
1878 use the upload.ConvergentUploadMixin class to achieve this
1881 Uploadables that do not care about convergence (or do not wish to
1882 make multiple passes over the data) can simply return a
1883 strongly-random 16 byte string.
1885 get_encryption_key() may be called multiple times: the IUploadable is
1886 required to return the same value each time.
1890 """Return a Deferred that fires with a list of strings (perhaps with
1891 only a single element) that, when concatenated together, contain the
1892 next 'length' bytes of data. If EOF is near, this may provide fewer
1893 than 'length' bytes. The total number of bytes provided by read()
1894 before it signals EOF must equal the size provided by get_size().
1896 If the data must be acquired through multiple internal read
1897 operations, returning a list instead of a single string may help to
1898 reduce string copies. However, the length of the concatenated strings
1899 must equal the amount of data requested, unless EOF is encountered.
1900 Long reads, or short reads without EOF, are not allowed. read()
1901 should return the same amount of data as a local disk file read, just
1902 in a different shape and asynchronously.
1904 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1905 so a 10kB file means length=3kB, 100kB file means length=30kB,
1906 and >=1MB file means length=300kB.
1908 This method provides for a single full pass through the data. Later
1909 use cases may desire multiple passes or access to only parts of the
1910 data (such as a mutable file making small edits-in-place). This API
1911 will be expanded once those use cases are better understood.
1915 """The upload is finished, and whatever filehandle was in use may be
1919 class IMutableUploadable(Interface):
1921 I represent content that is due to be uploaded to a mutable filecap.
1923 # This is somewhat simpler than the IUploadable interface above
1924 # because mutable files do not need to be concerned with possibly
1925 # generating a CHK, nor with per-file keys. It is a subset of the
1926 # methods in IUploadable, though, so we could just as well implement
1927 # the mutable uploadables as IUploadables that don't happen to use
1928 # those methods (with the understanding that the unused methods will
1929 # never be called on such objects)
1932 Returns a Deferred that fires with the size of the content held
1938 Returns a list of strings that, when concatenated, are the next
1939 length bytes of the file, or fewer if there are fewer bytes
1940 between the current location and the end of the file.
1945 The process that used the Uploadable is finished using it, so
1946 the uploadable may be closed.
1950 class IUploadResults(Interface):
1951 """I am returned by immutable upload() methods and contain the results of
1954 Note that some of my methods return empty values (0 or an empty dict)
1955 when called for non-distributed LIT files."""
1957 def get_file_size():
1958 """Return the file size, in bytes."""
1961 """Return the (string) URI of the object uploaded, a CHK readcap."""
1963 def get_ciphertext_fetched():
1964 """Return the number of bytes fetched by the helpe for this upload,
1965 or 0 if the helper did not need to fetch any bytes (or if there was
1968 def get_preexisting_shares():
1969 """Return the number of shares that were already present in the grid."""
1971 def get_pushed_shares():
1972 """Return the number of shares that were uploaded."""
1975 """Return a dict mapping share identifier to set of IServer
1976 instances. This indicates which servers were given which shares. For
1977 immutable files, the shareid is an integer (the share number, from 0
1978 to N-1). For mutable files, it is a string of the form
1979 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
1980 the share number."""
1982 def get_servermap():
1983 """Return dict mapping IServer instance to a set of share numbers."""
1986 """Return dict of timing information, mapping name to seconds. All
1988 total : total upload time, start to finish
1989 storage_index : time to compute the storage index
1990 peer_selection : time to decide which peers will be used
1991 contacting_helper : initial helper query to upload/no-upload decision
1992 helper_total : initial helper query to helper finished pushing
1993 cumulative_fetch : helper waiting for ciphertext requests
1994 total_fetch : helper start to last ciphertext response
1995 cumulative_encoding : just time spent in zfec
1996 cumulative_sending : just time spent waiting for storage servers
1997 hashes_and_close : last segment push to shareholder close
1998 total_encode_and_push : first encode to shareholder close
2001 def get_uri_extension_data():
2002 """Return the dict of UEB data created for this file."""
2004 def get_verifycapstr():
2005 """Return the (string) verify-cap URI for the uploaded object."""
2008 class IDownloadResults(Interface):
2009 """I am created internally by download() methods. I contain a number of
2010 public attributes that contain details about the download process.::
2012 .file_size : the size of the file, in bytes
2013 .servers_used : set of server peerids that were used during download
2014 .server_problems : dict mapping server peerid to a problem string. Only
2015 servers that had problems (bad hashes, disconnects)
2017 .servermap : dict mapping server peerid to a set of share numbers. Only
2018 servers that had any shares are listed here.
2019 .timings : dict of timing information, mapping name to seconds (float)
2020 peer_selection : time to ask servers about shares
2021 servers_peer_selection : dict of peerid to DYHB-query time
2022 uri_extension : time to fetch a copy of the URI extension block
2023 hashtrees : time to fetch the hash trees
2024 segments : time to fetch, decode, and deliver segments
2025 cumulative_fetch : time spent waiting for storage servers
2026 cumulative_decode : just time spent in zfec
2027 cumulative_decrypt : just time spent in decryption
2028 total : total download time, start to finish
2029 fetch_per_server : dict of server to list of per-segment fetch times
2033 class IUploader(Interface):
2034 def upload(uploadable):
2035 """Upload the file. 'uploadable' must impement IUploadable. This
2036 returns a Deferred that fires with an IUploadResults instance, from
2037 which the URI of the file can be obtained as results.uri ."""
2040 class ICheckable(Interface):
2041 def check(monitor, verify=False, add_lease=False):
2042 """Check up on my health, optionally repairing any problems.
2044 This returns a Deferred that fires with an instance that provides
2045 ICheckResults, or None if the object is non-distributed (i.e. LIT
2048 The monitor will be checked periodically to see if the operation has
2049 been cancelled. If so, no new queries will be sent, and the Deferred
2050 will fire (with a OperationCancelledError) immediately.
2052 Filenodes and dirnodes (which provide IFilesystemNode) are also
2053 checkable. Instances that represent verifier-caps will be checkable
2054 but not downloadable. Some objects (like LIT files) do not actually
2055 live in the grid, and their checkers return None (non-distributed
2056 files are always healthy).
2058 If verify=False, a relatively lightweight check will be performed: I
2059 will ask all servers if they have a share for me, and I will believe
2060 whatever they say. If there are at least N distinct shares on the
2061 grid, my results will indicate r.is_healthy()==True. This requires a
2062 roundtrip to each server, but does not transfer very much data, so
2063 the network bandwidth is fairly low.
2065 If verify=True, a more resource-intensive check will be performed:
2066 every share will be downloaded, and the hashes will be validated on
2067 every bit. I will ignore any shares that failed their hash checks. If
2068 there are at least N distinct valid shares on the grid, my results
2069 will indicate r.is_healthy()==True. This requires N/k times as much
2070 download bandwidth (and server disk IO) as a regular download. If a
2071 storage server is holding a corrupt share, or is experiencing memory
2072 failures during retrieval, or is malicious or buggy, then
2073 verification will detect the problem, but checking will not.
2075 If add_lease=True, I will ensure that an up-to-date lease is present
2076 on each share. The lease secrets will be derived from by node secret
2077 (in BASEDIR/private/secret), so either I will add a new lease to the
2078 share, or I will merely renew the lease that I already had. In a
2079 future version of the storage-server protocol (once Accounting has
2080 been implemented), there may be additional options here to define the
2081 kind of lease that is obtained (which account number to claim, etc).
2083 TODO: any problems seen during checking will be reported to the
2084 health-manager.furl, a centralized object that is responsible for
2085 figuring out why files are unhealthy so corrective action can be
2089 def check_and_repair(monitor, verify=False, add_lease=False):
2090 """Like check(), but if the file/directory is not healthy, attempt to
2093 Any non-healthy result will cause an immediate repair operation, to
2094 generate and upload new shares. After repair, the file will be as
2095 healthy as we can make it. Details about what sort of repair is done
2096 will be put in the check-and-repair results. The Deferred will not
2097 fire until the repair is complete.
2099 This returns a Deferred that fires with an instance of
2100 ICheckAndRepairResults."""
2103 class IDeepCheckable(Interface):
2104 def start_deep_check(verify=False, add_lease=False):
2105 """Check upon the health of me and everything I can reach.
2107 This is a recursive form of check(), useable only on dirnodes.
2109 I return a Monitor, with results that are an IDeepCheckResults
2112 TODO: If any of the directories I traverse are unrecoverable, the
2113 Monitor will report failure. If any of the files I check upon are
2114 unrecoverable, those problems will be reported in the
2115 IDeepCheckResults as usual, and the Monitor will not report a
2119 def start_deep_check_and_repair(verify=False, add_lease=False):
2120 """Check upon the health of me and everything I can reach. Repair
2121 anything that isn't healthy.
2123 This is a recursive form of check_and_repair(), useable only on
2126 I return a Monitor, with results that are an
2127 IDeepCheckAndRepairResults object.
2129 TODO: If any of the directories I traverse are unrecoverable, the
2130 Monitor will report failure. If any of the files I check upon are
2131 unrecoverable, those problems will be reported in the
2132 IDeepCheckResults as usual, and the Monitor will not report a
2137 class ICheckResults(Interface):
2138 """I contain the detailed results of a check/verify operation.
2141 def get_storage_index():
2142 """Return a string with the (binary) storage index."""
2144 def get_storage_index_string():
2145 """Return a string with the (printable) abbreviated storage index."""
2148 """Return the (string) URI of the object that was checked."""
2151 """Return a boolean, True if the file/dir is fully healthy, False if
2152 it is damaged in any way. Non-distributed LIT files always return
2155 def is_recoverable():
2156 """Return a boolean, True if the file/dir can be recovered, False if
2157 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2158 files always return True."""
2160 # the following methods all return None for non-distributed LIT files
2162 def get_happiness():
2163 """Return the happiness count of the file."""
2165 def get_encoding_needed():
2166 """Return 'k', the number of shares required for recovery."""
2168 def get_encoding_expected():
2169 """Return 'N', the number of total shares generated."""
2171 def get_share_counter_good():
2172 """Return the number of distinct good shares that were found. For
2173 mutable files, this counts shares for the 'best' version."""
2175 def get_share_counter_wrong():
2176 """For mutable files, return the number of shares for versions other
2177 than the 'best' one (which is defined as being the recoverable
2178 version with the highest sequence number, then the highest roothash).
2179 These are either leftover shares from an older version (perhaps on a
2180 server that was offline when an update occurred), shares from an
2181 unrecoverable newer version, or shares from an alternate current
2182 version that results from an uncoordinated write collision. For a
2183 healthy file, this will equal 0. For immutable files, this will
2186 def get_corrupt_shares():
2187 """Return a list of 'share locators', one for each share that was
2188 found to be corrupt (integrity failure). Each share locator is a list
2189 of (IServer, storage_index, sharenum)."""
2191 def get_incompatible_shares():
2192 """Return a list of 'share locators', one for each share that was
2193 found to be of an unknown format. Each share locator is a list of
2194 (IServer, storage_index, sharenum)."""
2196 def get_servers_responding():
2197 """Return a list of IServer objects, one for each server that
2198 responded to the share query (even if they said they didn't have
2199 shares, and even if they said they did have shares but then didn't
2200 send them when asked, or dropped the connection, or returned a
2201 Failure, and even if they said they did have shares and sent
2202 incorrect ones when asked)"""
2204 def get_host_counter_good_shares():
2205 """Return the number of distinct storage servers with good shares. If
2206 this number is less than get_share_counters()[good], then some shares
2207 are doubled up, increasing the correlation of failures. This
2208 indicates that one or more shares should be moved to an otherwise
2209 unused server, if one is available.
2212 def get_version_counter_recoverable():
2213 """Return the number of recoverable versions of the file. For a
2214 healthy file, this will equal 1."""
2216 def get_version_counter_unrecoverable():
2217 """Return the number of unrecoverable versions of the file. For a
2218 healthy file, this will be 0."""
2221 """Return a dict mapping share identifier to list of IServer objects.
2222 This indicates which servers are holding which shares. For immutable
2223 files, the shareid is an integer (the share number, from 0 to N-1).
2224 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2225 containing the sequence number, the roothash, and the share number."""
2228 """Return a string with a brief (one-line) summary of the results."""
2231 """Return a list of strings with more detailed results."""
2234 class ICheckAndRepairResults(Interface):
2235 """I contain the detailed results of a check/verify/repair operation.
2237 The IFilesystemNode.check()/verify()/repair() methods all return
2238 instances that provide ICheckAndRepairResults.
2241 def get_storage_index():
2242 """Return a string with the (binary) storage index."""
2244 def get_storage_index_string():
2245 """Return a string with the (printable) abbreviated storage index."""
2247 def get_repair_attempted():
2248 """Return a boolean, True if a repair was attempted. We might not
2249 attempt to repair the file because it was healthy, or healthy enough
2250 (i.e. some shares were missing but not enough to exceed some
2251 threshold), or because we don't know how to repair this object."""
2253 def get_repair_successful():
2254 """Return a boolean, True if repair was attempted and the file/dir
2255 was fully healthy afterwards. False if no repair was attempted or if
2256 a repair attempt failed."""
2258 def get_pre_repair_results():
2259 """Return an ICheckResults instance that describes the state of the
2260 file/dir before any repair was attempted."""
2262 def get_post_repair_results():
2263 """Return an ICheckResults instance that describes the state of the
2264 file/dir after any repair was attempted. If no repair was attempted,
2265 the pre-repair and post-repair results will be identical."""
2268 class IDeepCheckResults(Interface):
2269 """I contain the results of a deep-check operation.
2271 This is returned by a call to ICheckable.deep_check().
2274 def get_root_storage_index_string():
2275 """Return the storage index (abbreviated human-readable string) of
2276 the first object checked."""
2279 """Return a dictionary with the following keys::
2281 count-objects-checked: count of how many objects were checked
2282 count-objects-healthy: how many of those objects were completely
2284 count-objects-unhealthy: how many were damaged in some way
2285 count-objects-unrecoverable: how many were unrecoverable
2286 count-corrupt-shares: how many shares were found to have
2287 corruption, summed over all objects
2291 def get_corrupt_shares():
2292 """Return a set of (IServer, storage_index, sharenum) for all shares
2293 that were found to be corrupt. storage_index is binary."""
2295 def get_all_results():
2296 """Return a dictionary mapping pathname (a tuple of strings, ready to
2297 be slash-joined) to an ICheckResults instance, one for each object
2298 that was checked."""
2300 def get_results_for_storage_index(storage_index):
2301 """Retrive the ICheckResults instance for the given (binary)
2302 storage index. Raises KeyError if there are no results for that
2306 """Return a dictionary with the same keys as
2307 IDirectoryNode.deep_stats()."""
2310 class IDeepCheckAndRepairResults(Interface):
2311 """I contain the results of a deep-check-and-repair operation.
2313 This is returned by a call to ICheckable.deep_check_and_repair().
2316 def get_root_storage_index_string():
2317 """Return the storage index (abbreviated human-readable string) of
2318 the first object checked."""
2321 """Return a dictionary with the following keys::
2323 count-objects-checked: count of how many objects were checked
2324 count-objects-healthy-pre-repair: how many of those objects were
2325 completely healthy (before any
2327 count-objects-unhealthy-pre-repair: how many were damaged in
2329 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2330 count-objects-healthy-post-repair: how many of those objects were
2331 completely healthy (after any
2333 count-objects-unhealthy-post-repair: how many were damaged in
2335 count-objects-unrecoverable-post-repair: how many were
2337 count-repairs-attempted: repairs were attempted on this many
2338 objects. The count-repairs- keys will
2339 always be provided, however unless
2340 repair=true is present, they will all
2342 count-repairs-successful: how many repairs resulted in healthy
2344 count-repairs-unsuccessful: how many repairs resulted did not
2345 results in completely healthy objects
2346 count-corrupt-shares-pre-repair: how many shares were found to
2347 have corruption, summed over all
2348 objects examined (before any
2350 count-corrupt-shares-post-repair: how many shares were found to
2351 have corruption, summed over all
2352 objects examined (after any
2357 """Return a dictionary with the same keys as
2358 IDirectoryNode.deep_stats()."""
2360 def get_corrupt_shares():
2361 """Return a set of (IServer, storage_index, sharenum) for all shares
2362 that were found to be corrupt before any repair was attempted.
2363 storage_index is binary.
2365 def get_remaining_corrupt_shares():
2366 """Return a set of (IServer, storage_index, sharenum) for all shares
2367 that were found to be corrupt after any repair was completed.
2368 storage_index is binary. These are shares that need manual inspection
2369 and probably deletion.
2371 def get_all_results():
2372 """Return a dictionary mapping pathname (a tuple of strings, ready to
2373 be slash-joined) to an ICheckAndRepairResults instance, one for each
2374 object that was checked."""
2376 def get_results_for_storage_index(storage_index):
2377 """Retrive the ICheckAndRepairResults instance for the given (binary)
2378 storage index. Raises KeyError if there are no results for that
2382 class IRepairable(Interface):
2383 def repair(check_results):
2384 """Attempt to repair the given object. Returns a Deferred that fires
2385 with a IRepairResults object.
2387 I must be called with an object that implements ICheckResults, as
2388 proof that you have actually discovered a problem with this file. I
2389 will use the data in the checker results to guide the repair process,
2390 such as which servers provided bad data and should therefore be
2391 avoided. The ICheckResults object is inside the
2392 ICheckAndRepairResults object, which is returned by the
2393 ICheckable.check() method::
2395 d = filenode.check(repair=False)
2396 def _got_results(check_and_repair_results):
2397 check_results = check_and_repair_results.get_pre_repair_results()
2398 return filenode.repair(check_results)
2399 d.addCallback(_got_results)
2404 class IRepairResults(Interface):
2405 """I contain the results of a repair operation."""
2406 def get_successful():
2407 """Returns a boolean: True if the repair made the file healthy, False
2408 if not. Repair failure generally indicates a file that has been
2409 damaged beyond repair."""
2412 class IClient(Interface):
2413 def upload(uploadable):
2414 """Upload some data into a CHK, get back the UploadResults for it.
2415 @param uploadable: something that implements IUploadable
2416 @return: a Deferred that fires with the UploadResults instance.
2417 To get the URI for this file, use results.uri .
2420 def create_mutable_file(contents=""):
2421 """Create a new mutable file (with initial) contents, get back the
2424 @param contents: (bytestring, callable, or None): this provides the
2425 initial contents of the mutable file. If 'contents' is a bytestring,
2426 it will be used as-is. If 'contents' is a callable, it will be
2427 invoked with the new MutableFileNode instance and is expected to
2428 return a bytestring with the initial contents of the file (the
2429 callable can use node.get_writekey() to decide how to encrypt the
2430 initial contents, e.g. for a brand new dirnode with initial
2431 children). contents=None is equivalent to an empty string. Using
2432 content_maker= is more efficient than creating a mutable file and
2433 setting its contents in two separate operations.
2435 @return: a Deferred that fires with an IMutableFileNode instance.
2438 def create_dirnode(initial_children={}):
2439 """Create a new unattached dirnode, possibly with initial children.
2441 @param initial_children: dict with keys that are unicode child names,
2442 and values that are (childnode, metadata) tuples.
2444 @return: a Deferred that fires with the new IDirectoryNode instance.
2447 def create_node_from_uri(uri, rouri):
2448 """Create a new IFilesystemNode instance from the uri, synchronously.
2449 @param uri: a string or IURI-providing instance, or None. This could
2450 be for a LiteralFileNode, a CHK file node, a mutable file
2451 node, or a directory node
2452 @param rouri: a string or IURI-providing instance, or None. If the
2453 main uri is None, I will use the rouri instead. If I
2454 recognize the format of the main uri, I will ignore the
2455 rouri (because it can be derived from the writecap).
2457 @return: an instance that provides IFilesystemNode (or more usefully
2458 one of its subclasses). File-specifying URIs will result in
2459 IFileNode-providing instances, like ImmutableFileNode,
2460 LiteralFileNode, or MutableFileNode. Directory-specifying
2461 URIs will result in IDirectoryNode-providing instances, like
2466 class INodeMaker(Interface):
2467 """The NodeMaker is used to create IFilesystemNode instances. It can
2468 accept a filecap/dircap string and return the node right away. It can
2469 also create new nodes (i.e. upload a file, or create a mutable file)
2470 asynchronously. Once you have one of these nodes, you can use other
2471 methods to determine whether it is a file or directory, and to download
2472 or modify its contents.
2474 The NodeMaker encapsulates all the authorities that these
2475 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2476 Tahoe process will typically have a single NodeMaker, but unit tests may
2477 create simplified/mocked forms for testing purposes.
2480 def create_from_cap(writecap, readcap=None, deep_immutable=False, name=u"<unknown name>"):
2481 """I create an IFilesystemNode from the given writecap/readcap. I can
2482 only provide nodes for existing file/directory objects: use my other
2483 methods to create new objects. I return synchronously."""
2485 def create_mutable_file(contents=None, keysize=None):
2486 """I create a new mutable file, and return a Deferred that will fire
2487 with the IMutableFileNode instance when it is ready. If contents= is
2488 provided (a bytestring), it will be used as the initial contents of
2489 the new file, otherwise the file will contain zero bytes. keysize= is
2490 for use by unit tests, to create mutable files that are smaller than
2493 def create_new_mutable_directory(initial_children={}):
2494 """I create a new mutable directory, and return a Deferred that will
2495 fire with the IDirectoryNode instance when it is ready. If
2496 initial_children= is provided (a dict mapping unicode child name to
2497 (childnode, metadata_dict) tuples), the directory will be populated
2498 with those children, otherwise it will be empty."""
2501 class IClientStatus(Interface):
2502 def list_all_uploads():
2503 """Return a list of uploader objects, one for each upload that
2504 currently has an object available (tracked with weakrefs). This is
2505 intended for debugging purposes."""
2507 def list_active_uploads():
2508 """Return a list of active IUploadStatus objects."""
2510 def list_recent_uploads():
2511 """Return a list of IUploadStatus objects for the most recently
2514 def list_all_downloads():
2515 """Return a list of downloader objects, one for each download that
2516 currently has an object available (tracked with weakrefs). This is
2517 intended for debugging purposes."""
2519 def list_active_downloads():
2520 """Return a list of active IDownloadStatus objects."""
2522 def list_recent_downloads():
2523 """Return a list of IDownloadStatus objects for the most recently
2524 started downloads."""
2527 class IUploadStatus(Interface):
2529 """Return a timestamp (float with seconds since epoch) indicating
2530 when the operation was started."""
2532 def get_storage_index():
2533 """Return a string with the (binary) storage index in use on this
2534 upload. Returns None if the storage index has not yet been
2538 """Return an integer with the number of bytes that will eventually
2539 be uploaded for this file. Returns None if the size is not yet known.
2542 """Return True if this upload is using a Helper, False if not."""
2545 """Return a string describing the current state of the upload
2549 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2550 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2551 made towards hashing the file to determine a CHK encryption key: if
2552 non-convergent encryption is in use, this will be trivial, otherwise
2553 the whole file must be hashed. 'ciphertext' describes how much of the
2554 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2555 uploads. 'encode_and_push' describes how much of the encode-and-push
2556 process has finished: for helper uploads this is dependent upon the
2557 helper providing progress reports. It might be reasonable to add all
2558 three numbers and report the sum to the user."""
2561 """Return True if the upload is currently active, False if not."""
2564 """Return an instance of UploadResults (which contains timing and
2565 sharemap information). Might return None if the upload is not yet
2569 """Each upload status gets a unique number: this method returns that
2570 number. This provides a handle to this particular upload, so a web
2571 page can generate a suitable hyperlink."""
2574 class IDownloadStatus(Interface):
2576 """Return a timestamp (float with seconds since epoch) indicating
2577 when the operation was started."""
2579 def get_storage_index():
2580 """Return a string with the (binary) storage index in use on this
2581 download. This may be None if there is no storage index (i.e. LIT
2585 """Return an integer with the number of bytes that will eventually be
2586 retrieved for this file. Returns None if the size is not yet known.
2590 """Return True if this download is using a Helper, False if not."""
2593 """Return a string describing the current state of the download
2597 """Returns a float (from 0.0 to 1.0) describing the amount of the
2598 download that has completed. This value will remain at 0.0 until the
2599 first byte of plaintext is pushed to the download target."""
2602 """Return True if the download is currently active, False if not."""
2605 """Each download status gets a unique number: this method returns
2606 that number. This provides a handle to this particular download, so a
2607 web page can generate a suitable hyperlink."""
2610 class IServermapUpdaterStatus(Interface):
2613 class IPublishStatus(Interface):
2616 class IRetrieveStatus(Interface):
2620 class NotCapableError(Exception):
2621 """You have tried to write to a read-only node."""
2623 class BadWriteEnablerError(Exception):
2627 class RIControlClient(RemoteInterface):
2628 def wait_for_client_connections(num_clients=int):
2629 """Do not return until we have connections to at least NUM_CLIENTS
2635 def upload_random_data_from_file(size=int, convergence=str):
2638 def download_to_tempfile_and_delete(uri=str):
2641 def get_memory_usage():
2642 """Return a dict describes the amount of memory currently in use. The
2643 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2644 measuring memory consupmtion in bytes."""
2645 return DictOf(str, int)
2647 def speed_test(count=int, size=int, mutable=Any()):
2648 """Write 'count' tempfiles to disk, all of the given size. Measure
2649 how long (in seconds) it takes to upload them all to the servers.
2650 Then measure how long it takes to download all of them. If 'mutable'
2651 is 'create', time creation of mutable files. If 'mutable' is
2652 'upload', then time access to the same mutable file instead of
2655 Returns a tuple of (upload_time, download_time).
2657 return (float, float)
2659 def measure_peer_response_time():
2660 """Send a short message to each connected peer, and measure the time
2661 it takes for them to respond to it. This is a rough measure of the
2662 application-level round trip time.
2664 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2667 return DictOf(str, float)
2670 UploadResults = Any() #DictOf(str, str)
2673 class RIEncryptedUploadable(RemoteInterface):
2674 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2679 def get_all_encoding_parameters():
2680 return (int, int, int, long)
2682 def read_encrypted(offset=Offset, length=ReadSize):
2689 class RICHKUploadHelper(RemoteInterface):
2690 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2694 Return a dictionary of version information.
2696 return DictOf(str, Any())
2698 def upload(reader=RIEncryptedUploadable):
2699 return UploadResults
2702 class RIHelper(RemoteInterface):
2703 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2707 Return a dictionary of version information.
2709 return DictOf(str, Any())
2711 def upload_chk(si=StorageIndex):
2712 """See if a file with a given storage index needs uploading. The
2713 helper will ask the appropriate storage servers to see if the file
2714 has already been uploaded. If so, the helper will return a set of
2715 'upload results' that includes whatever hashes are needed to build
2716 the read-cap, and perhaps a truncated sharemap.
2718 If the file has not yet been uploaded (or if it was only partially
2719 uploaded), the helper will return an empty upload-results dictionary
2720 and also an RICHKUploadHelper object that will take care of the
2721 upload process. The client should call upload() on this object and
2722 pass it a reference to an RIEncryptedUploadable object that will
2723 provide ciphertext. When the upload is finished, the upload() method
2724 will finish and return the upload results.
2726 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2729 class RIStatsProvider(RemoteInterface):
2730 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2732 Provides access to statistics and monitoring information.
2737 returns a dictionary containing 'counters' and 'stats', each a
2738 dictionary with string counter/stat name keys, and numeric or None values.
2739 counters are monotonically increasing measures of work done, and
2740 stats are instantaneous measures (potentially time averaged
2743 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2746 class RIStatsGatherer(RemoteInterface):
2747 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2749 Provides a monitoring service for centralised collection of stats
2752 def provide(provider=RIStatsProvider, nickname=str):
2754 @param provider: a stats collector instance that should be polled
2755 periodically by the gatherer to collect stats.
2756 @param nickname: a name useful to identify the provided client
2761 class IStatsProducer(Interface):
2764 returns a dictionary, with str keys representing the names of stats
2765 to be monitored, and numeric values.
2768 class RIKeyGenerator(RemoteInterface):
2769 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2771 Provides a service offering to make RSA key pairs.
2774 def get_rsa_key_pair(key_size=int):
2776 @param key_size: the size of the signature key.
2777 @return: tuple(verifying_key, signing_key)
2779 return TupleOf(str, str)
2782 class FileTooLargeError(Exception):
2786 class IValidatedThingProxy(Interface):
2788 """ Acquire a thing and validate it. Return a deferred that is
2789 eventually fired with self if the thing is valid or errbacked if it
2790 can't be acquired or validated."""
2793 class InsufficientVersionError(Exception):
2794 def __init__(self, needed, got):
2795 self.needed = needed
2799 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2802 class EmptyPathnameComponentError(Exception):
2803 """The webapi disallows empty pathname components."""