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 last_connect_time: when we last established a connection
421 last_loss_time: when we last lost a connection
423 version: the peer's version, from the most recent connection
424 oldest_supported: the peer's oldest supported version, same
426 rref: the RemoteReference, if connected, otherwise None
427 remote_host: the IAddress, if connected, otherwise None
429 This method is intended for monitoring interfaces, such as a web page
430 that describes connecting and connected peers.
433 def get_all_peerids():
434 """Return a frozenset of all peerids to whom we have a connection (to
435 one or more services) established. Mostly useful for unit tests."""
437 def get_all_connections_for(service_name):
438 """Return a frozenset of (nodeid, service_name, rref) tuples, one
439 for each active connection that provides the given SERVICE_NAME."""
441 def get_permuted_peers(service_name, key):
442 """Returns an ordered list of (peerid, rref) tuples, selecting from
443 the connections that provide SERVICE_NAME, using a hash-based
444 permutation keyed by KEY. This randomizes the service list in a
445 repeatable way, to distribute load over many peers.
449 class IDisplayableServer(Interface):
460 class IServer(IDisplayableServer):
461 """I live in the client, and represent a single server."""
462 def start_connecting(tub, trigger_cb):
466 """Once a server is connected, I return a RemoteReference.
467 Before a server is connected for the first time, I return None.
469 Note that the rref I return will start producing DeadReferenceErrors
470 once the connection is lost.
474 class IMutableSlotWriter(Interface):
476 The interface for a writer around a mutable slot on a remote server.
478 def set_checkstring(seqnum_or_checkstring, root_hash=None, salt=None):
480 Set the checkstring that I will pass to the remote server when
483 @param checkstring A packed checkstring to use.
485 Note that implementations can differ in which semantics they
486 wish to support for set_checkstring -- they can, for example,
487 build the checkstring themselves from its constituents, or
491 def get_checkstring():
493 Get the checkstring that I think currently exists on the remote
497 def put_block(data, segnum, salt):
499 Add a block and salt to the share.
502 def put_encprivkey(encprivkey):
504 Add the encrypted private key to the share.
507 def put_blockhashes(blockhashes):
509 @param blockhashes=list
510 Add the block hash tree to the share.
513 def put_sharehashes(sharehashes):
515 @param sharehashes=dict
516 Add the share hash chain to the share.
521 Return the part of the share that needs to be signed.
524 def put_signature(signature):
526 Add the signature to the share.
529 def put_verification_key(verification_key):
531 Add the verification key to the share.
534 def finish_publishing():
536 Do anything necessary to finish writing the share to a remote
537 server. I require that no further publishing needs to take place
538 after this method has been called.
542 class IURI(Interface):
543 def init_from_string(uri):
544 """Accept a string (as created by my to_string() method) and populate
545 this instance with its data. I am not normally called directly,
546 please use the module-level uri.from_string() function to convert
547 arbitrary URI strings into IURI-providing instances."""
550 """Return False if this URI be used to modify the data. Return True
551 if this URI cannot be used to modify the data."""
554 """Return True if the data can be modified by *somebody* (perhaps
555 someone who has a more powerful URI than this one)."""
557 # TODO: rename to get_read_cap()
559 """Return another IURI instance that represents a read-only form of
560 this one. If is_readonly() is True, this returns self."""
562 def get_verify_cap():
563 """Return an instance that provides IVerifierURI, which can be used
564 to check on the availability of the file or directory, without
565 providing enough capabilities to actually read or modify the
566 contents. This may return None if the file does not need checking or
567 verification (e.g. LIT URIs).
571 """Return a string of printable ASCII characters, suitable for
572 passing into init_from_string."""
575 class IVerifierURI(Interface, IURI):
576 def init_from_string(uri):
577 """Accept a string (as created by my to_string() method) and populate
578 this instance with its data. I am not normally called directly,
579 please use the module-level uri.from_string() function to convert
580 arbitrary URI strings into IURI-providing instances."""
583 """Return a string of printable ASCII characters, suitable for
584 passing into init_from_string."""
587 class IDirnodeURI(Interface):
588 """I am a URI that represents a dirnode."""
591 class IFileURI(Interface):
592 """I am a URI that represents a filenode."""
594 """Return the length (in bytes) of the file that I represent."""
597 class IImmutableFileURI(IFileURI):
600 class IMutableFileURI(Interface):
603 class IDirectoryURI(Interface):
606 class IReadonlyDirectoryURI(Interface):
610 class CapConstraintError(Exception):
611 """A constraint on a cap was violated."""
613 class MustBeDeepImmutableError(CapConstraintError):
614 """Mutable children cannot be added to an immutable directory.
615 Also, caps obtained from an immutable directory can trigger this error
616 if they are later found to refer to a mutable object and then used."""
618 class MustBeReadonlyError(CapConstraintError):
619 """Known write caps cannot be specified in a ro_uri field. Also,
620 caps obtained from a ro_uri field can trigger this error if they
621 are later found to be write caps and then used."""
623 class MustNotBeUnknownRWError(CapConstraintError):
624 """Cannot add an unknown child cap specified in a rw_uri field."""
627 class IReadable(Interface):
628 """I represent a readable object -- either an immutable file, or a
629 specific version of a mutable file.
633 """Return True if this reference provides mutable access to the given
634 file or directory (i.e. if you can modify it), or False if not. Note
635 that even if this reference is read-only, someone else may hold a
636 read-write reference to it.
638 For an IReadable returned by get_best_readable_version(), this will
639 always return True, but for instances of subinterfaces such as
640 IMutableFileVersion, it may return False."""
643 """Return True if this file or directory is mutable (by *somebody*,
644 not necessarily you), False if it is is immutable. Note that a file
645 might be mutable overall, but your reference to it might be
646 read-only. On the other hand, all references to an immutable file
647 will be read-only; there are no read-write references to an immutable
650 def get_storage_index():
651 """Return the storage index of the file."""
654 """Return the length (in bytes) of this readable object."""
656 def download_to_data():
657 """Download all of the file contents. I return a Deferred that fires
658 with the contents as a byte string."""
660 def read(consumer, offset=0, size=None):
661 """Download a portion (possibly all) of the file's contents, making
662 them available to the given IConsumer. Return a Deferred that fires
663 (with the consumer) when the consumer is unregistered (either because
664 the last byte has been given to it, or because the consumer threw an
665 exception during write(), possibly because it no longer wants to
666 receive data). The portion downloaded will start at 'offset' and
667 contain 'size' bytes (or the remainder of the file if size==None). It
668 is an error to read beyond the end of the file: callers must use
669 get_size() and clip any non-default offset= and size= parameters. It
670 is permissible to read zero bytes.
672 The consumer will be used in non-streaming mode: an IPullProducer
673 will be attached to it.
675 The consumer will not receive data right away: several network trips
676 must occur first. The order of events will be::
678 consumer.registerProducer(p, streaming)
679 (if streaming == False)::
680 consumer does p.resumeProducing()
682 consumer does p.resumeProducing()
683 consumer.write(data).. (repeat until all data is written)
684 consumer.unregisterProducer()
685 deferred.callback(consumer)
687 If a download error occurs, or an exception is raised by
688 consumer.registerProducer() or consumer.write(), I will call
689 consumer.unregisterProducer() and then deliver the exception via
690 deferred.errback(). To cancel the download, the consumer should call
691 p.stopProducing(), which will result in an exception being delivered
692 via deferred.errback().
694 See src/allmydata/util/consumer.py for an example of a simple
695 download-to-memory consumer.
699 class IWriteable(Interface):
701 I define methods that callers can use to update SDMF and MDMF
702 mutable files on a Tahoe-LAFS grid.
704 # XXX: For the moment, we have only this. It is possible that we
705 # want to move overwrite() and modify() in here too.
706 def update(data, offset):
708 I write the data from my data argument to the MDMF file,
709 starting at offset. I continue writing data until my data
710 argument is exhausted, appending data to the file as necessary.
712 # assert IMutableUploadable.providedBy(data)
713 # to append data: offset=node.get_size_of_best_version()
714 # do we want to support compacting MDMF?
715 # for an MDMF file, this can be done with O(data.get_size())
716 # memory. For an SDMF file, any modification takes
717 # O(node.get_size_of_best_version()).
720 class IMutableFileVersion(IReadable):
721 """I provide access to a particular version of a mutable file. The
722 access is read/write if I was obtained from a filenode derived from
723 a write cap, or read-only if the filenode was derived from a read cap.
726 def get_sequence_number():
727 """Return the sequence number of this version."""
730 """Return the IMutableFileServerMap instance that was used to create
735 """Return this filenode's writekey, or None if the node does not have
736 write-capability. This may be used to assist with data structures
737 that need to make certain data available only to writers, such as the
738 read-write child caps in dirnodes. The recommended process is to have
739 reader-visible data be submitted to the filenode in the clear (where
740 it will be encrypted by the filenode using the readkey), but encrypt
741 writer-visible data using this writekey.
744 def overwrite(new_contents):
745 """Replace the contents of the mutable file, provided that no other
746 node has published (or is attempting to publish, concurrently) a
747 newer version of the file than this one.
749 I will avoid modifying any share that is different than the version
750 given by get_sequence_number(). However, if another node is writing
751 to the file at the same time as me, I may manage to update some shares
752 while they update others. If I see any evidence of this, I will signal
753 UncoordinatedWriteError, and the file will be left in an inconsistent
754 state (possibly the version you provided, possibly the old version,
755 possibly somebody else's version, and possibly a mix of shares from
758 The recommended response to UncoordinatedWriteError is to either
759 return it to the caller (since they failed to coordinate their
760 writes), or to attempt some sort of recovery. It may be sufficient to
761 wait a random interval (with exponential backoff) and repeat your
762 operation. If I do not signal UncoordinatedWriteError, then I was
763 able to write the new version without incident.
765 I return a Deferred that fires (with a PublishStatus object) when the
766 update has completed.
769 def modify(modifier_cb):
770 """Modify the contents of the file, by downloading this version,
771 applying the modifier function (or bound method), then uploading
772 the new version. This will succeed as long as no other node
773 publishes a version between the download and the upload.
774 I return a Deferred that fires (with a PublishStatus object) when
775 the update is complete.
777 The modifier callable will be given three arguments: a string (with
778 the old contents), a 'first_time' boolean, and a servermap. As with
779 download_to_data(), the old contents will be from this version,
780 but the modifier can use the servermap to make other decisions
781 (such as refusing to apply the delta if there are multiple parallel
782 versions, or if there is evidence of a newer unrecoverable version).
783 'first_time' will be True the first time the modifier is called,
784 and False on any subsequent calls.
786 The callable should return a string with the new contents. The
787 callable must be prepared to be called multiple times, and must
788 examine the input string to see if the change that it wants to make
789 is already present in the old version. If it does not need to make
790 any changes, it can either return None, or return its input string.
792 If the modifier raises an exception, it will be returned in the
797 # The hierarchy looks like this:
804 class IFilesystemNode(Interface):
806 """Return the strongest 'cap instance' associated with this node.
807 (writecap for writeable-mutable files/directories, readcap for
808 immutable or readonly-mutable files/directories). To convert this
809 into a string, call .to_string() on the result."""
812 """Return a readonly cap instance for this node. For immutable or
813 readonly nodes, get_cap() and get_readcap() return the same thing."""
815 def get_repair_cap():
816 """Return an IURI instance that can be used to repair the file, or
817 None if this node cannot be repaired (either because it is not
818 distributed, like a LIT file, or because the node does not represent
819 sufficient authority to create a repair-cap, like a read-only RSA
820 mutable file node [which cannot create the correct write-enablers]).
823 def get_verify_cap():
824 """Return an IVerifierURI instance that represents the
825 'verifiy/refresh capability' for this node. The holder of this
826 capability will be able to renew the lease for this node, protecting
827 it from garbage-collection. They will also be able to ask a server if
828 it holds a share for the file or directory.
832 """Return the URI string corresponding to the strongest cap associated
833 with this node. If this node is read-only, the URI will only offer
834 read-only access. If this node is read-write, the URI will offer
837 If you have read-write access to a node and wish to share merely
838 read-only access with others, use get_readonly_uri().
842 """Return the URI string that can be used by others to get write
843 access to this node, if it is writeable. If this is a read-only node,
846 def get_readonly_uri():
847 """Return the URI string that can be used by others to get read-only
848 access to this node. The result is a read-only URI, regardless of
849 whether this node is read-only or read-write.
851 If you have merely read-only access to this node, get_readonly_uri()
852 will return the same thing as get_uri().
855 def get_storage_index():
856 """Return a string with the (binary) storage index in use on this
857 download. This may be None if there is no storage index (i.e. LIT
858 files and directories)."""
861 """Return True if this reference provides mutable access to the given
862 file or directory (i.e. if you can modify it), or False if not. Note
863 that even if this reference is read-only, someone else may hold a
864 read-write reference to it."""
867 """Return True if this file or directory is mutable (by *somebody*,
868 not necessarily you), False if it is is immutable. Note that a file
869 might be mutable overall, but your reference to it might be
870 read-only. On the other hand, all references to an immutable file
871 will be read-only; there are no read-write references to an immutable
876 """Return True if this is an unknown node."""
878 def is_allowed_in_immutable_directory():
879 """Return True if this node is allowed as a child of a deep-immutable
880 directory. This is true if either the node is of a known-immutable type,
881 or it is unknown and read-only.
885 """Raise any error associated with this node."""
887 # XXX: These may not be appropriate outside the context of an IReadable.
889 """Return the length (in bytes) of the data this node represents. For
890 directory nodes, I return the size of the backing store. I return
891 synchronously and do not consult the network, so for mutable objects,
892 I will return the most recently observed size for the object, or None
893 if I don't remember a size. Use get_current_size, which returns a
894 Deferred, if you want more up-to-date information."""
896 def get_current_size():
897 """I return a Deferred that fires with the length (in bytes) of the
898 data this node represents.
902 class IFileNode(IFilesystemNode):
903 """I am a node that represents a file: a sequence of bytes. I am not a
904 container, like IDirectoryNode."""
905 def get_best_readable_version():
906 """Return a Deferred that fires with an IReadable for the 'best'
907 available version of the file. The IReadable provides only read
908 access, even if this filenode was derived from a write cap.
910 For an immutable file, there is only one version. For a mutable
911 file, the 'best' version is the recoverable version with the
912 highest sequence number. If no uncoordinated writes have occurred,
913 and if enough shares are available, then this will be the most
914 recent version that has been uploaded. If no version is recoverable,
915 the Deferred will errback with an UnrecoverableFileError.
918 def download_best_version():
919 """Download the contents of the version that would be returned
920 by get_best_readable_version(). This is equivalent to calling
921 download_to_data() on the IReadable given by that method.
923 I return a Deferred that fires with a byte string when the file
924 has been fully downloaded. To support streaming download, use
925 the 'read' method of IReadable. If no version is recoverable,
926 the Deferred will errback with an UnrecoverableFileError.
929 def get_size_of_best_version():
930 """Find the size of the version that would be returned by
931 get_best_readable_version().
933 I return a Deferred that fires with an integer. If no version
934 is recoverable, the Deferred will errback with an
935 UnrecoverableFileError.
939 class IImmutableFileNode(IFileNode, IReadable):
940 """I am a node representing an immutable file. Immutable files have
944 class IMutableFileNode(IFileNode):
945 """I provide access to a 'mutable file', which retains its identity
946 regardless of what contents are put in it.
948 The consistency-vs-availability problem means that there might be
949 multiple versions of a file present in the grid, some of which might be
950 unrecoverable (i.e. have fewer than 'k' shares). These versions are
951 loosely ordered: each has a sequence number and a hash, and any version
952 with seqnum=N was uploaded by a node that has seen at least one version
955 The 'servermap' (an instance of IMutableFileServerMap) is used to
956 describe the versions that are known to be present in the grid, and which
957 servers are hosting their shares. It is used to represent the 'state of
958 the world', and is used for this purpose by my test-and-set operations.
959 Downloading the contents of the mutable file will also return a
960 servermap. Uploading a new version into the mutable file requires a
961 servermap as input, and the semantics of the replace operation is
962 'replace the file with my new version if it looks like nobody else has
963 changed the file since my previous download'. Because the file is
964 distributed, this is not a perfect test-and-set operation, but it will do
965 its best. If the replace process sees evidence of a simultaneous write,
966 it will signal an UncoordinatedWriteError, so that the caller can take
970 Most readers will want to use the 'best' current version of the file, and
971 should use my 'download_best_version()' method.
973 To unconditionally replace the file, callers should use overwrite(). This
974 is the mode that user-visible mutable files will probably use.
976 To apply some delta to the file, call modify() with a callable modifier
977 function that can apply the modification that you want to make. This is
978 the mode that dirnodes will use, since most directory modification
979 operations can be expressed in terms of deltas to the directory state.
982 Three methods are available for users who need to perform more complex
983 operations. The first is get_servermap(), which returns an up-to-date
984 servermap using a specified mode. The second is download_version(), which
985 downloads a specific version (not necessarily the 'best' one). The third
986 is 'upload', which accepts new contents and a servermap (which must have
987 been updated with MODE_WRITE). The upload method will attempt to apply
988 the new contents as long as no other node has modified the file since the
989 servermap was updated. This might be useful to a caller who wants to
990 merge multiple versions into a single new one.
992 Note that each time the servermap is updated, a specific 'mode' is used,
993 which determines how many peers are queried. To use a servermap for my
994 replace() method, that servermap must have been updated in MODE_WRITE.
995 These modes are defined in allmydata.mutable.common, and consist of
996 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
997 allmydata/mutable/servermap.py for details about the differences.
999 Mutable files are currently limited in size (about 3.5MB max) and can
1000 only be retrieved and updated all-at-once, as a single big string. Future
1001 versions of our mutable files will remove this restriction.
1003 def get_best_mutable_version():
1004 """Return a Deferred that fires with an IMutableFileVersion for
1005 the 'best' available version of the file. The best version is
1006 the recoverable version with the highest sequence number. If no
1007 uncoordinated writes have occurred, and if enough shares are
1008 available, then this will be the most recent version that has
1011 If no version is recoverable, the Deferred will errback with an
1012 UnrecoverableFileError.
1015 def overwrite(new_contents):
1016 """Unconditionally replace the contents of the mutable file with new
1017 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
1018 is only appropriate to use when the new contents of the file are
1019 completely unrelated to the old ones, and you do not care about other
1022 I return a Deferred that fires (with a PublishStatus object) when the
1023 update has completed.
1026 def modify(modifier_cb):
1027 """Modify the contents of the file, by downloading the current
1028 version, applying the modifier function (or bound method), then
1029 uploading the new version. I return a Deferred that fires (with a
1030 PublishStatus object) when the update is complete.
1032 The modifier callable will be given three arguments: a string (with
1033 the old contents), a 'first_time' boolean, and a servermap. As with
1034 download_best_version(), the old contents will be from the best
1035 recoverable version, but the modifier can use the servermap to make
1036 other decisions (such as refusing to apply the delta if there are
1037 multiple parallel versions, or if there is evidence of a newer
1038 unrecoverable version). 'first_time' will be True the first time the
1039 modifier is called, and False on any subsequent calls.
1041 The callable should return a string with the new contents. The
1042 callable must be prepared to be called multiple times, and must
1043 examine the input string to see if the change that it wants to make
1044 is already present in the old version. If it does not need to make
1045 any changes, it can either return None, or return its input string.
1047 If the modifier raises an exception, it will be returned in the
1051 def get_servermap(mode):
1052 """Return a Deferred that fires with an IMutableFileServerMap
1053 instance, updated using the given mode.
1056 def download_version(servermap, version):
1057 """Download a specific version of the file, using the servermap
1058 as a guide to where the shares are located.
1060 I return a Deferred that fires with the requested contents, or
1061 errbacks with UnrecoverableFileError. Note that a servermap that was
1062 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1063 all versions (those modes stop querying servers as soon as they can
1064 fulfil their goals), so you may want to use MODE_CHECK (which checks
1065 everything) to get increased visibility.
1068 def upload(new_contents, servermap):
1069 """Replace the contents of the file with new ones. This requires a
1070 servermap that was previously updated with MODE_WRITE.
1072 I attempt to provide test-and-set semantics, in that I will avoid
1073 modifying any share that is different than the version I saw in the
1074 servermap. However, if another node is writing to the file at the
1075 same time as me, I may manage to update some shares while they update
1076 others. If I see any evidence of this, I will signal
1077 UncoordinatedWriteError, and the file will be left in an inconsistent
1078 state (possibly the version you provided, possibly the old version,
1079 possibly somebody else's version, and possibly a mix of shares from
1082 The recommended response to UncoordinatedWriteError is to either
1083 return it to the caller (since they failed to coordinate their
1084 writes), or to attempt some sort of recovery. It may be sufficient to
1085 wait a random interval (with exponential backoff) and repeat your
1086 operation. If I do not signal UncoordinatedWriteError, then I was
1087 able to write the new version without incident.
1089 I return a Deferred that fires (with a PublishStatus object) when the
1090 publish has completed. I will update the servermap in-place with the
1091 location of all new shares.
1095 """Return this filenode's writekey, or None if the node does not have
1096 write-capability. This may be used to assist with data structures
1097 that need to make certain data available only to writers, such as the
1098 read-write child caps in dirnodes. The recommended process is to have
1099 reader-visible data be submitted to the filenode in the clear (where
1100 it will be encrypted by the filenode using the readkey), but encrypt
1101 writer-visible data using this writekey.
1105 """Returns the mutable file protocol version."""
1108 class NotEnoughSharesError(Exception):
1109 """Download was unable to get enough shares"""
1111 class NoSharesError(Exception):
1112 """Download was unable to get any shares at all."""
1114 class DownloadStopped(Exception):
1117 class UploadUnhappinessError(Exception):
1118 """Upload was unable to satisfy 'servers_of_happiness'"""
1120 class UnableToFetchCriticalDownloadDataError(Exception):
1121 """I was unable to fetch some piece of critical data that is supposed to
1122 be identically present in all shares."""
1124 class NoServersError(Exception):
1125 """Upload wasn't given any servers to work with, usually indicating a
1126 network or Introducer problem."""
1128 class ExistingChildError(Exception):
1129 """A directory node was asked to add or replace a child that already
1130 exists, and overwrite= was set to False."""
1132 class NoSuchChildError(Exception):
1133 """A directory node was asked to fetch a child that does not exist."""
1135 # avoid UnicodeEncodeErrors when converting to str
1136 return self.__repr__()
1138 class ChildOfWrongTypeError(Exception):
1139 """An operation was attempted on a child of the wrong type (file or directory)."""
1142 class IDirectoryNode(IFilesystemNode):
1143 """I represent a filesystem node that is a container, with a
1144 name-to-child mapping, holding the tahoe equivalent of a directory. All
1145 child names are unicode strings, and all children are some sort of
1146 IFilesystemNode (a file, subdirectory, or unknown node).
1151 The dirnode ('1') URI returned by this method can be used in
1152 set_uri() on a different directory ('2') to 'mount' a reference to
1153 this directory ('1') under the other ('2'). This URI is just a
1154 string, so it can be passed around through email or other out-of-band
1158 def get_readonly_uri():
1160 The dirnode ('1') URI returned by this method can be used in
1161 set_uri() on a different directory ('2') to 'mount' a reference to
1162 this directory ('1') under the other ('2'). This URI is just a
1163 string, so it can be passed around through email or other out-of-band
1168 """I return a Deferred that fires with a dictionary mapping child
1169 name (a unicode string) to (node, metadata_dict) tuples, in which
1170 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1173 def has_child(name):
1174 """I return a Deferred that fires with a boolean, True if there
1175 exists a child of the given name, False if not. The child name must
1176 be a unicode string."""
1179 """I return a Deferred that fires with a specific named child node,
1180 which is an IFilesystemNode. The child name must be a unicode string.
1181 I raise NoSuchChildError if I do not have a child by that name."""
1183 def get_metadata_for(name):
1184 """I return a Deferred that fires with the metadata dictionary for
1185 a specific named child node. The child name must be a unicode string.
1186 This metadata is stored in the *edge*, not in the child, so it is
1187 attached to the parent dirnode rather than the child node.
1188 I raise NoSuchChildError if I do not have a child by that name."""
1190 def set_metadata_for(name, metadata):
1191 """I replace any existing metadata for the named child with the new
1192 metadata. The child name must be a unicode string. This metadata is
1193 stored in the *edge*, not in the child, so it is attached to the
1194 parent dirnode rather than the child node. I return a Deferred
1195 (that fires with this dirnode) when the operation is complete.
1196 I raise NoSuchChildError if I do not have a child by that name."""
1198 def get_child_at_path(path):
1199 """Transform a child path into an IFilesystemNode.
1201 I perform a recursive series of 'get' operations to find the named
1202 descendant node. I return a Deferred that fires with the node, or
1203 errbacks with NoSuchChildError if the node could not be found.
1205 The path can be either a single string (slash-separated) or a list of
1206 path-name elements. All elements must be unicode strings.
1209 def get_child_and_metadata_at_path(path):
1210 """Transform a child path into an IFilesystemNode and metadata.
1212 I am like get_child_at_path(), but my Deferred fires with a tuple of
1213 (node, metadata). The metadata comes from the last edge. If the path
1214 is empty, the metadata will be an empty dictionary.
1217 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1218 """I add a child (by writecap+readcap) at the specific name. I return
1219 a Deferred that fires when the operation finishes. If overwrite= is
1220 True, I will replace any existing child of the same name, otherwise
1221 an existing child will cause me to return ExistingChildError. The
1222 child name must be a unicode string.
1224 The child caps could be for a file, or for a directory. If you have
1225 both the writecap and readcap, you should provide both arguments.
1226 If you have only one cap and don't know whether it is read-only,
1227 provide it as the writecap argument and leave the readcap as None.
1228 If you have only one cap that is known to be read-only, provide it
1229 as the readcap argument and leave the writecap as None.
1230 The filecaps are typically obtained from an IFilesystemNode with
1231 get_uri() and get_readonly_uri().
1233 If metadata= is provided, I will use it as the metadata for the named
1234 edge. This will replace any existing metadata. If metadata= is left
1235 as the default value of None, I will set ['mtime'] to the current
1236 time, and I will set ['ctime'] to the current time if there was not
1237 already a child by this name present. This roughly matches the
1238 ctime/mtime semantics of traditional filesystems. See the
1239 "About the metadata" section of webapi.txt for futher information.
1241 If this directory node is read-only, the Deferred will errback with a
1242 NotWriteableError."""
1244 def set_children(entries, overwrite=True):
1245 """Add multiple children (by writecap+readcap) to a directory node.
1246 Takes a dictionary, with childname as keys and (writecap, readcap)
1247 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1248 a Deferred that fires (with this dirnode) when the operation
1249 finishes. This is equivalent to calling set_uri() multiple times, but
1250 is much more efficient. All child names must be unicode strings.
1253 def set_node(name, child, metadata=None, overwrite=True):
1254 """I add a child at the specific name. I return a Deferred that fires
1255 when the operation finishes. This Deferred will fire with the child
1256 node that was just added. I will replace any existing child of the
1257 same name. The child name must be a unicode string. The 'child'
1258 instance must be an instance providing IFilesystemNode.
1260 If metadata= is provided, I will use it as the metadata for the named
1261 edge. This will replace any existing metadata. If metadata= is left
1262 as the default value of None, I will set ['mtime'] to the current
1263 time, and I will set ['ctime'] to the current time if there was not
1264 already a child by this name present. This roughly matches the
1265 ctime/mtime semantics of traditional filesystems. See the
1266 "About the metadata" section of webapi.txt for futher information.
1268 If this directory node is read-only, the Deferred will errback with a
1269 NotWriteableError."""
1271 def set_nodes(entries, overwrite=True):
1272 """Add multiple children to a directory node. Takes a dict mapping
1273 unicode childname to (child_node, metdata) tuples. If metdata=None,
1274 the original metadata is left unmodified. Returns a Deferred that
1275 fires (with this dirnode) when the operation finishes. This is
1276 equivalent to calling set_node() multiple times, but is much more
1279 def add_file(name, uploadable, metadata=None, overwrite=True):
1280 """I upload a file (using the given IUploadable), then attach the
1281 resulting ImmutableFileNode to the directory at the given name. I set
1282 metadata the same way as set_uri and set_node. The child name must be
1285 I return a Deferred that fires (with the IFileNode of the uploaded
1286 file) when the operation completes."""
1288 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1289 """I remove the child at the specific name. I return a Deferred that
1290 fires when the operation finishes. The child name must be a unicode
1291 string. If must_exist is True and I do not have a child by that name,
1292 I raise NoSuchChildError. If must_be_directory is True and the child
1293 is a file, or if must_be_file is True and the child is a directory,
1294 I raise ChildOfWrongTypeError."""
1296 def create_subdirectory(name, initial_children={}, overwrite=True,
1297 mutable=True, mutable_version=None, metadata=None):
1298 """I create and attach a directory at the given name. The new
1299 directory can be empty, or it can be populated with children
1300 according to 'initial_children', which takes a dictionary in the same
1301 format as set_nodes (i.e. mapping unicode child name to (childnode,
1302 metadata) tuples). The child name must be a unicode string. I return
1303 a Deferred that fires (with the new directory node) when the
1304 operation finishes."""
1306 def move_child_to(current_child_name, new_parent, new_child_name=None,
1308 """I take one of my children and move them to a new parent. The child
1309 is referenced by name. On the new parent, the child will live under
1310 'new_child_name', which defaults to 'current_child_name'. TODO: what
1311 should we do about metadata? I return a Deferred that fires when the
1312 operation finishes. The child name must be a unicode string. I raise
1313 NoSuchChildError if I do not have a child by that name."""
1315 def build_manifest():
1316 """I generate a table of everything reachable from this directory.
1317 I also compute deep-stats as described below.
1319 I return a Monitor. The Monitor's results will be a dictionary with
1322 res['manifest']: a list of (path, cap) tuples for all nodes
1323 (directories and files) reachable from this one.
1324 'path' will be a tuple of unicode strings. The
1325 origin dirnode will be represented by an empty path
1327 res['verifycaps']: a list of (printable) verifycap strings, one for
1328 each reachable non-LIT node. This is a set:
1329 it will contain no duplicates.
1330 res['storage-index']: a list of (base32) storage index strings,
1331 one for each reachable non-LIT node. This is
1332 a set: it will contain no duplicates.
1333 res['stats']: a dictionary, the same that is generated by
1334 start_deep_stats() below.
1336 The Monitor will also have an .origin_si attribute with the (binary)
1337 storage index of the starting point.
1340 def start_deep_stats():
1341 """Return a Monitor, examining all nodes (directories and files)
1342 reachable from this one. The Monitor's results will be a dictionary
1343 with the following keys::
1345 count-immutable-files: count of how many CHK files are in the set
1346 count-mutable-files: same, for mutable files (does not include
1348 count-literal-files: same, for LIT files
1349 count-files: sum of the above three
1351 count-directories: count of directories
1353 size-immutable-files: total bytes for all CHK files in the set
1354 size-mutable-files (TODO): same, for current version of all mutable
1355 files, does not include directories
1356 size-literal-files: same, for LIT files
1357 size-directories: size of mutable files used by directories
1359 largest-directory: number of bytes in the largest directory
1360 largest-directory-children: number of children in the largest
1362 largest-immutable-file: number of bytes in the largest CHK file
1364 size-mutable-files is not yet implemented, because it would involve
1365 even more queries than deep_stats does.
1367 The Monitor will also have an .origin_si attribute with the (binary)
1368 storage index of the starting point.
1370 This operation will visit every directory node underneath this one,
1371 and can take a long time to run. On a typical workstation with good
1372 bandwidth, this can examine roughly 15 directories per second (and
1373 takes several minutes of 100% CPU for ~1700 directories).
1377 class ICodecEncoder(Interface):
1378 def set_params(data_size, required_shares, max_shares):
1379 """Set up the parameters of this encoder.
1381 This prepares the encoder to perform an operation that converts a
1382 single block of data into a number of shares, such that a future
1383 ICodecDecoder can use a subset of these shares to recover the
1384 original data. This operation is invoked by calling encode(). Once
1385 the encoding parameters are set up, the encode operation can be
1386 invoked multiple times.
1388 set_params() prepares the encoder to accept blocks of input data that
1389 are exactly 'data_size' bytes in length. The encoder will be prepared
1390 to produce 'max_shares' shares for each encode() operation (although
1391 see the 'desired_share_ids' to use less CPU). The encoding math will
1392 be chosen such that the decoder can get by with as few as
1393 'required_shares' of these shares and still reproduce the original
1394 data. For example, set_params(1000, 5, 5) offers no redundancy at
1395 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1397 Numerical Restrictions: 'data_size' is required to be an integral
1398 multiple of 'required_shares'. In general, the caller should choose
1399 required_shares and max_shares based upon their reliability
1400 requirements and the number of peers available (the total storage
1401 space used is roughly equal to max_shares*data_size/required_shares),
1402 then choose data_size to achieve the memory footprint desired (larger
1403 data_size means more efficient operation, smaller data_size means
1404 smaller memory footprint).
1406 In addition, 'max_shares' must be equal to or greater than
1407 'required_shares'. Of course, setting them to be equal causes
1408 encode() to degenerate into a particularly slow form of the 'split'
1411 See encode() for more details about how these parameters are used.
1413 set_params() must be called before any other ICodecEncoder methods
1418 """Return the 3-tuple of data_size, required_shares, max_shares"""
1420 def get_encoder_type():
1421 """Return a short string that describes the type of this encoder.
1423 There is required to be a global table of encoder classes. This method
1424 returns an index into this table; the value at this index is an
1425 encoder class, and this encoder is an instance of that class.
1428 def get_block_size():
1429 """Return the length of the shares that encode() will produce.
1432 def encode_proposal(data, desired_share_ids=None):
1433 """Encode some data.
1435 'data' must be a string (or other buffer object), and len(data) must
1436 be equal to the 'data_size' value passed earlier to set_params().
1438 This will return a Deferred that will fire with two lists. The first
1439 is a list of shares, each of which is a string (or other buffer
1440 object) such that len(share) is the same as what get_share_size()
1441 returned earlier. The second is a list of shareids, in which each is
1442 an integer. The lengths of the two lists will always be equal to each
1443 other. The user should take care to keep each share closely
1444 associated with its shareid, as one is useless without the other.
1446 The length of this output list will normally be the same as the value
1447 provided to the 'max_shares' parameter of set_params(). This may be
1448 different if 'desired_share_ids' is provided.
1450 'desired_share_ids', if provided, is required to be a sequence of
1451 ints, each of which is required to be >= 0 and < max_shares. If not
1452 provided, encode() will produce 'max_shares' shares, as if
1453 'desired_share_ids' were set to range(max_shares). You might use this
1454 if you initially thought you were going to use 10 peers, started
1455 encoding, and then two of the peers dropped out: you could use
1456 desired_share_ids= to skip the work (both memory and CPU) of
1457 producing shares for the peers that are no longer available.
1461 def encode(inshares, desired_share_ids=None):
1462 """Encode some data. This may be called multiple times. Each call is
1465 inshares is a sequence of length required_shares, containing buffers
1466 (i.e. strings), where each buffer contains the next contiguous
1467 non-overlapping segment of the input data. Each buffer is required to
1468 be the same length, and the sum of the lengths of the buffers is
1469 required to be exactly the data_size promised by set_params(). (This
1470 implies that the data has to be padded before being passed to
1471 encode(), unless of course it already happens to be an even multiple
1472 of required_shares in length.)
1474 Note: the requirement to break up your data into
1475 'required_shares' chunks of exactly the right length before
1476 calling encode() is surprising from point of view of a user
1477 who doesn't know how FEC works. It feels like an
1478 implementation detail that has leaked outside the abstraction
1479 barrier. Is there a use case in which the data to be encoded
1480 might already be available in pre-segmented chunks, such that
1481 it is faster or less work to make encode() take a list rather
1482 than splitting a single string?
1484 Yes, there is: suppose you are uploading a file with K=64,
1485 N=128, segsize=262,144. Then each in-share will be of size
1486 4096. If you use this .encode() API then your code could first
1487 read each successive 4096-byte chunk from the file and store
1488 each one in a Python string and store each such Python string
1489 in a Python list. Then you could call .encode(), passing that
1490 list as "inshares". The encoder would generate the other 64
1491 "secondary shares" and return to you a new list containing
1492 references to the same 64 Python strings that you passed in
1493 (as the primary shares) plus references to the new 64 Python
1496 (You could even imagine that your code could use readv() so
1497 that the operating system can arrange to get all of those
1498 bytes copied from the file into the Python list of Python
1499 strings as efficiently as possible instead of having a loop
1500 written in C or in Python to copy the next part of the file
1501 into the next string.)
1503 On the other hand if you instead use the .encode_proposal()
1504 API (above), then your code can first read in all of the
1505 262,144 bytes of the segment from the file into a Python
1506 string, then call .encode_proposal() passing the segment data
1507 as the "data" argument. The encoder would basically first
1508 split the "data" argument into a list of 64 in-shares of 4096
1509 byte each, and then do the same thing that .encode() does. So
1510 this would result in a little bit more copying of data and a
1511 little bit higher of a "maximum memory usage" during the
1512 process, although it might or might not make a practical
1513 difference for our current use cases.
1515 Note that "inshares" is a strange name for the parameter if
1516 you think of the parameter as being just for feeding in data
1517 to the codec. It makes more sense if you think of the result
1518 of this encoding as being the set of shares from inshares plus
1519 an extra set of "secondary shares" (or "check shares"). It is
1520 a surprising name! If the API is going to be surprising then
1521 the name should be surprising. If we switch to
1522 encode_proposal() above then we should also switch to an
1525 'desired_share_ids', if provided, is required to be a sequence of
1526 ints, each of which is required to be >= 0 and < max_shares. If not
1527 provided, encode() will produce 'max_shares' shares, as if
1528 'desired_share_ids' were set to range(max_shares). You might use this
1529 if you initially thought you were going to use 10 peers, started
1530 encoding, and then two of the peers dropped out: you could use
1531 desired_share_ids= to skip the work (both memory and CPU) of
1532 producing shares for the peers that are no longer available.
1534 For each call, encode() will return a Deferred that fires with two
1535 lists, one containing shares and the other containing the shareids.
1536 The get_share_size() method can be used to determine the length of
1537 the share strings returned by encode(). Each shareid is a small
1538 integer, exactly as passed into 'desired_share_ids' (or
1539 range(max_shares), if desired_share_ids was not provided).
1541 The shares and their corresponding shareids are required to be kept
1542 together during storage and retrieval. Specifically, the share data is
1543 useless by itself: the decoder needs to be told which share is which
1544 by providing it with both the shareid and the actual share data.
1546 This function will allocate an amount of memory roughly equal to::
1548 (max_shares - required_shares) * get_share_size()
1550 When combined with the memory that the caller must allocate to
1551 provide the input data, this leads to a memory footprint roughly
1552 equal to the size of the resulting encoded shares (i.e. the expansion
1553 factor times the size of the input segment).
1558 # returning a list of (shareidN,shareN) tuples instead of a pair of
1559 # lists (shareids..,shares..). Brian thought the tuples would
1560 # encourage users to keep the share and shareid together throughout
1561 # later processing, Zooko pointed out that the code to iterate
1562 # through two lists is not really more complicated than using a list
1563 # of tuples and there's also a performance improvement
1565 # having 'data_size' not required to be an integral multiple of
1566 # 'required_shares'. Doing this would require encode() to perform
1567 # padding internally, and we'd prefer to have any padding be done
1568 # explicitly by the caller. Yes, it is an abstraction leak, but
1569 # hopefully not an onerous one.
1572 class ICodecDecoder(Interface):
1573 def set_params(data_size, required_shares, max_shares):
1574 """Set the params. They have to be exactly the same ones that were
1575 used for encoding."""
1577 def get_needed_shares():
1578 """Return the number of shares needed to reconstruct the data.
1579 set_params() is required to be called before this."""
1581 def decode(some_shares, their_shareids):
1582 """Decode a partial list of shares into data.
1584 'some_shares' is required to be a sequence of buffers of sharedata, a
1585 subset of the shares returned by ICodecEncode.encode(). Each share is
1586 required to be of the same length. The i'th element of their_shareids
1587 is required to be the shareid of the i'th buffer in some_shares.
1589 This returns a Deferred that fires with a sequence of buffers. This
1590 sequence will contain all of the segments of the original data, in
1591 order. The sum of the lengths of all of the buffers will be the
1592 'data_size' value passed into the original ICodecEncode.set_params()
1593 call. To get back the single original input block of data, use
1594 ''.join(output_buffers), or you may wish to simply write them in
1595 order to an output file.
1597 Note that some of the elements in the result sequence may be
1598 references to the elements of the some_shares input sequence. In
1599 particular, this means that if those share objects are mutable (e.g.
1600 arrays) and if they are changed, then both the input (the
1601 'some_shares' parameter) and the output (the value given when the
1602 deferred is triggered) will change.
1604 The length of 'some_shares' is required to be exactly the value of
1605 'required_shares' passed into the original ICodecEncode.set_params()
1610 class IEncoder(Interface):
1611 """I take an object that provides IEncryptedUploadable, which provides
1612 encrypted data, and a list of shareholders. I then encode, hash, and
1613 deliver shares to those shareholders. I will compute all the necessary
1614 Merkle hash trees that are necessary to validate the crypttext that
1615 eventually comes back from the shareholders. I provide the URI Extension
1616 Block Hash, and the encoding parameters, both of which must be included
1619 I do not choose shareholders, that is left to the IUploader. I must be
1620 given a dict of RemoteReferences to storage buckets that are ready and
1621 willing to receive data.
1625 """Specify the number of bytes that will be encoded. This must be
1626 peformed before get_serialized_params() can be called.
1629 def set_encrypted_uploadable(u):
1630 """Provide a source of encrypted upload data. 'u' must implement
1631 IEncryptedUploadable.
1633 When this is called, the IEncryptedUploadable will be queried for its
1634 length and the storage_index that should be used.
1636 This returns a Deferred that fires with this Encoder instance.
1638 This must be performed before start() can be called.
1641 def get_param(name):
1642 """Return an encoding parameter, by name.
1644 'storage_index': return a string with the (16-byte truncated SHA-256
1645 hash) storage index to which these shares should be
1648 'share_counts': return a tuple describing how many shares are used:
1649 (needed_shares, servers_of_happiness, total_shares)
1651 'num_segments': return an int with the number of segments that
1654 'segment_size': return an int with the size of each segment.
1656 'block_size': return the size of the individual blocks that will
1657 be delivered to a shareholder's put_block() method. By
1658 knowing this, the shareholder will be able to keep all
1659 blocks in a single file and still provide random access
1660 when reading them. # TODO: can we avoid exposing this?
1662 'share_size': an int with the size of the data that will be stored
1663 on each shareholder. This is aggregate amount of data
1664 that will be sent to the shareholder, summed over all
1665 the put_block() calls I will ever make. It is useful to
1666 determine this size before asking potential
1667 shareholders whether they will grant a lease or not,
1668 since their answers will depend upon how much space we
1669 need. TODO: this might also include some amount of
1670 overhead, like the size of all the hashes. We need to
1671 decide whether this is useful or not.
1673 'serialized_params': a string with a concise description of the
1674 codec name and its parameters. This may be passed
1675 into the IUploadable to let it make sure that
1676 the same file encoded with different parameters
1677 will result in different storage indexes.
1679 Once this is called, set_size() and set_params() may not be called.
1682 def set_shareholders(shareholders, servermap):
1683 """Tell the encoder where to put the encoded shares. 'shareholders'
1684 must be a dictionary that maps share number (an integer ranging from
1685 0 to n-1) to an instance that provides IStorageBucketWriter.
1686 'servermap' is a dictionary that maps share number (as defined above)
1687 to a set of peerids. This must be performed before start() can be
1691 """Begin the encode/upload process. This involves reading encrypted
1692 data from the IEncryptedUploadable, encoding it, uploading the shares
1693 to the shareholders, then sending the hash trees.
1695 set_encrypted_uploadable() and set_shareholders() must be called
1696 before this can be invoked.
1698 This returns a Deferred that fires with a verify cap when the upload
1699 process is complete. The verifycap, plus the encryption key, is
1700 sufficient to construct the read cap.
1704 class IDecoder(Interface):
1705 """I take a list of shareholders and some setup information, then
1706 download, validate, decode, and decrypt data from them, writing the
1707 results to an output file.
1709 I do not locate the shareholders, that is left to the IDownloader. I must
1710 be given a dict of RemoteReferences to storage buckets that are ready to
1715 """I take a file-like object (providing write and close) to which all
1716 the plaintext data will be written.
1718 TODO: producer/consumer . Maybe write() should return a Deferred that
1719 indicates when it will accept more data? But probably having the
1720 IDecoder be a producer is easier to glue to IConsumer pieces.
1723 def set_shareholders(shareholders):
1724 """I take a dictionary that maps share identifiers (small integers)
1725 to RemoteReferences that provide RIBucketReader. This must be called
1729 """I start the download. This process involves retrieving data and
1730 hash chains from the shareholders, using the hashes to validate the
1731 data, decoding the shares into segments, decrypting the segments,
1732 then writing the resulting plaintext to the output file.
1734 I return a Deferred that will fire (with self) when the download is
1739 class IDownloadTarget(Interface):
1740 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1741 # will register itself as a producer. This allows the target to invoke
1742 # downloader.pauseProducing, resumeProducing, and stopProducing.
1744 """Called before any calls to write() or close(). If an error
1745 occurs before any data is available, fail() may be called without
1746 a previous call to open().
1748 'size' is the length of the file being downloaded, in bytes."""
1751 """Output some data to the target."""
1754 """Inform the target that there is no more data to be written."""
1757 """fail() is called to indicate that the download has failed. 'why'
1758 is a Failure object indicating what went wrong. No further methods
1759 will be invoked on the IDownloadTarget after fail()."""
1761 def register_canceller(cb):
1762 """The CiphertextDownloader uses this to register a no-argument function
1763 that the target can call to cancel the download. Once this canceller
1764 is invoked, no further calls to write() or close() will be made."""
1767 """When the CiphertextDownloader is done, this finish() function will be
1768 called. Whatever it returns will be returned to the invoker of
1769 Downloader.download.
1773 class IDownloader(Interface):
1774 def download(uri, target):
1775 """Perform a CHK download, sending the data to the given target.
1776 'target' must provide IDownloadTarget.
1778 Returns a Deferred that fires (with the results of target.finish)
1779 when the download is finished, or errbacks if something went wrong."""
1782 class IEncryptedUploadable(Interface):
1783 def set_upload_status(upload_status):
1784 """Provide an IUploadStatus object that should be filled with status
1785 information. The IEncryptedUploadable is responsible for setting
1786 key-determination progress ('chk'), size, storage_index, and
1787 ciphertext-fetch progress. It may delegate some of this
1788 responsibility to others, in particular to the IUploadable."""
1791 """This behaves just like IUploadable.get_size()."""
1793 def get_all_encoding_parameters():
1794 """Return a Deferred that fires with a tuple of
1795 (k,happy,n,segment_size). The segment_size will be used as-is, and
1796 must match the following constraints: it must be a multiple of k, and
1797 it shouldn't be unreasonably larger than the file size (if
1798 segment_size is larger than filesize, the difference must be stored
1801 This usually passes through to the IUploadable method of the same
1804 The encoder strictly obeys the values returned by this method. To
1805 make an upload use non-default encoding parameters, you must arrange
1806 to control the values that this method returns.
1809 def get_storage_index():
1810 """Return a Deferred that fires with a 16-byte storage index.
1813 def read_encrypted(length, hash_only):
1814 """This behaves just like IUploadable.read(), but returns crypttext
1815 instead of plaintext. If hash_only is True, then this discards the
1816 data (and returns an empty list); this improves efficiency when
1817 resuming an interrupted upload (where we need to compute the
1818 plaintext hashes, but don't need the redundant encrypted data)."""
1821 """Just like IUploadable.close()."""
1824 class IUploadable(Interface):
1825 def set_upload_status(upload_status):
1826 """Provide an IUploadStatus object that should be filled with status
1827 information. The IUploadable is responsible for setting
1828 key-determination progress ('chk')."""
1830 def set_default_encoding_parameters(params):
1831 """Set the default encoding parameters, which must be a dict mapping
1832 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1833 'max_segment_size'. These might have an influence on the final
1834 encoding parameters returned by get_all_encoding_parameters(), if the
1835 Uploadable doesn't have more specific preferences.
1837 This call is optional: if it is not used, the Uploadable will use
1838 some built-in defaults. If used, this method must be called before
1839 any other IUploadable methods to have any effect.
1843 """Return a Deferred that will fire with the length of the data to be
1844 uploaded, in bytes. This will be called before the data is actually
1845 used, to compute encoding parameters.
1848 def get_all_encoding_parameters():
1849 """Return a Deferred that fires with a tuple of
1850 (k,happy,n,segment_size). The segment_size will be used as-is, and
1851 must match the following constraints: it must be a multiple of k, and
1852 it shouldn't be unreasonably larger than the file size (if
1853 segment_size is larger than filesize, the difference must be stored
1856 The relative values of k and n allow some IUploadables to request
1857 better redundancy than others (in exchange for consuming more space
1860 Larger values of segment_size reduce hash overhead, while smaller
1861 values reduce memory footprint and cause data to be delivered in
1862 smaller pieces (which may provide a smoother and more predictable
1863 download experience).
1865 The encoder strictly obeys the values returned by this method. To
1866 make an upload use non-default encoding parameters, you must arrange
1867 to control the values that this method returns. One way to influence
1868 them may be to call set_encoding_parameters() before calling
1869 get_all_encoding_parameters().
1872 def get_encryption_key():
1873 """Return a Deferred that fires with a 16-byte AES key. This key will
1874 be used to encrypt the data. The key will also be hashed to derive
1877 Uploadables that want to achieve convergence should hash their file
1878 contents and the serialized_encoding_parameters to form the key
1879 (which of course requires a full pass over the data). Uploadables can
1880 use the upload.ConvergentUploadMixin class to achieve this
1883 Uploadables that do not care about convergence (or do not wish to
1884 make multiple passes over the data) can simply return a
1885 strongly-random 16 byte string.
1887 get_encryption_key() may be called multiple times: the IUploadable is
1888 required to return the same value each time.
1892 """Return a Deferred that fires with a list of strings (perhaps with
1893 only a single element) that, when concatenated together, contain the
1894 next 'length' bytes of data. If EOF is near, this may provide fewer
1895 than 'length' bytes. The total number of bytes provided by read()
1896 before it signals EOF must equal the size provided by get_size().
1898 If the data must be acquired through multiple internal read
1899 operations, returning a list instead of a single string may help to
1900 reduce string copies. However, the length of the concatenated strings
1901 must equal the amount of data requested, unless EOF is encountered.
1902 Long reads, or short reads without EOF, are not allowed. read()
1903 should return the same amount of data as a local disk file read, just
1904 in a different shape and asynchronously.
1906 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1907 so a 10kB file means length=3kB, 100kB file means length=30kB,
1908 and >=1MB file means length=300kB.
1910 This method provides for a single full pass through the data. Later
1911 use cases may desire multiple passes or access to only parts of the
1912 data (such as a mutable file making small edits-in-place). This API
1913 will be expanded once those use cases are better understood.
1917 """The upload is finished, and whatever filehandle was in use may be
1921 class IMutableUploadable(Interface):
1923 I represent content that is due to be uploaded to a mutable filecap.
1925 # This is somewhat simpler than the IUploadable interface above
1926 # because mutable files do not need to be concerned with possibly
1927 # generating a CHK, nor with per-file keys. It is a subset of the
1928 # methods in IUploadable, though, so we could just as well implement
1929 # the mutable uploadables as IUploadables that don't happen to use
1930 # those methods (with the understanding that the unused methods will
1931 # never be called on such objects)
1934 Returns a Deferred that fires with the size of the content held
1940 Returns a list of strings that, when concatenated, are the next
1941 length bytes of the file, or fewer if there are fewer bytes
1942 between the current location and the end of the file.
1947 The process that used the Uploadable is finished using it, so
1948 the uploadable may be closed.
1952 class IUploadResults(Interface):
1953 """I am returned by immutable upload() methods and contain the results of
1956 Note that some of my methods return empty values (0 or an empty dict)
1957 when called for non-distributed LIT files."""
1959 def get_file_size():
1960 """Return the file size, in bytes."""
1963 """Return the (string) URI of the object uploaded, a CHK readcap."""
1965 def get_ciphertext_fetched():
1966 """Return the number of bytes fetched by the helpe for this upload,
1967 or 0 if the helper did not need to fetch any bytes (or if there was
1970 def get_preexisting_shares():
1971 """Return the number of shares that were already present in the grid."""
1973 def get_pushed_shares():
1974 """Return the number of shares that were uploaded."""
1977 """Return a dict mapping share identifier to set of IServer
1978 instances. This indicates which servers were given which shares. For
1979 immutable files, the shareid is an integer (the share number, from 0
1980 to N-1). For mutable files, it is a string of the form
1981 'seq%d-%s-sh%d', containing the sequence number, the roothash, and
1982 the share number."""
1984 def get_servermap():
1985 """Return dict mapping IServer instance to a set of share numbers."""
1988 """Return dict of timing information, mapping name to seconds. All
1990 total : total upload time, start to finish
1991 storage_index : time to compute the storage index
1992 peer_selection : time to decide which peers will be used
1993 contacting_helper : initial helper query to upload/no-upload decision
1994 helper_total : initial helper query to helper finished pushing
1995 cumulative_fetch : helper waiting for ciphertext requests
1996 total_fetch : helper start to last ciphertext response
1997 cumulative_encoding : just time spent in zfec
1998 cumulative_sending : just time spent waiting for storage servers
1999 hashes_and_close : last segment push to shareholder close
2000 total_encode_and_push : first encode to shareholder close
2003 def get_uri_extension_data():
2004 """Return the dict of UEB data created for this file."""
2006 def get_verifycapstr():
2007 """Return the (string) verify-cap URI for the uploaded object."""
2010 class IDownloadResults(Interface):
2011 """I am created internally by download() methods. I contain a number of
2012 public attributes that contain details about the download process.::
2014 .file_size : the size of the file, in bytes
2015 .servers_used : set of server peerids that were used during download
2016 .server_problems : dict mapping server peerid to a problem string. Only
2017 servers that had problems (bad hashes, disconnects)
2019 .servermap : dict mapping server peerid to a set of share numbers. Only
2020 servers that had any shares are listed here.
2021 .timings : dict of timing information, mapping name to seconds (float)
2022 peer_selection : time to ask servers about shares
2023 servers_peer_selection : dict of peerid to DYHB-query time
2024 uri_extension : time to fetch a copy of the URI extension block
2025 hashtrees : time to fetch the hash trees
2026 segments : time to fetch, decode, and deliver segments
2027 cumulative_fetch : time spent waiting for storage servers
2028 cumulative_decode : just time spent in zfec
2029 cumulative_decrypt : just time spent in decryption
2030 total : total download time, start to finish
2031 fetch_per_server : dict of server to list of per-segment fetch times
2035 class IUploader(Interface):
2036 def upload(uploadable):
2037 """Upload the file. 'uploadable' must impement IUploadable. This
2038 returns a Deferred that fires with an IUploadResults instance, from
2039 which the URI of the file can be obtained as results.uri ."""
2042 class ICheckable(Interface):
2043 def check(monitor, verify=False, add_lease=False):
2044 """Check up on my health, optionally repairing any problems.
2046 This returns a Deferred that fires with an instance that provides
2047 ICheckResults, or None if the object is non-distributed (i.e. LIT
2050 The monitor will be checked periodically to see if the operation has
2051 been cancelled. If so, no new queries will be sent, and the Deferred
2052 will fire (with a OperationCancelledError) immediately.
2054 Filenodes and dirnodes (which provide IFilesystemNode) are also
2055 checkable. Instances that represent verifier-caps will be checkable
2056 but not downloadable. Some objects (like LIT files) do not actually
2057 live in the grid, and their checkers return None (non-distributed
2058 files are always healthy).
2060 If verify=False, a relatively lightweight check will be performed: I
2061 will ask all servers if they have a share for me, and I will believe
2062 whatever they say. If there are at least N distinct shares on the
2063 grid, my results will indicate r.is_healthy()==True. This requires a
2064 roundtrip to each server, but does not transfer very much data, so
2065 the network bandwidth is fairly low.
2067 If verify=True, a more resource-intensive check will be performed:
2068 every share will be downloaded, and the hashes will be validated on
2069 every bit. I will ignore any shares that failed their hash checks. If
2070 there are at least N distinct valid shares on the grid, my results
2071 will indicate r.is_healthy()==True. This requires N/k times as much
2072 download bandwidth (and server disk IO) as a regular download. If a
2073 storage server is holding a corrupt share, or is experiencing memory
2074 failures during retrieval, or is malicious or buggy, then
2075 verification will detect the problem, but checking will not.
2077 If add_lease=True, I will ensure that an up-to-date lease is present
2078 on each share. The lease secrets will be derived from by node secret
2079 (in BASEDIR/private/secret), so either I will add a new lease to the
2080 share, or I will merely renew the lease that I already had. In a
2081 future version of the storage-server protocol (once Accounting has
2082 been implemented), there may be additional options here to define the
2083 kind of lease that is obtained (which account number to claim, etc).
2085 TODO: any problems seen during checking will be reported to the
2086 health-manager.furl, a centralized object that is responsible for
2087 figuring out why files are unhealthy so corrective action can be
2091 def check_and_repair(monitor, verify=False, add_lease=False):
2092 """Like check(), but if the file/directory is not healthy, attempt to
2095 Any non-healthy result will cause an immediate repair operation, to
2096 generate and upload new shares. After repair, the file will be as
2097 healthy as we can make it. Details about what sort of repair is done
2098 will be put in the check-and-repair results. The Deferred will not
2099 fire until the repair is complete.
2101 This returns a Deferred that fires with an instance of
2102 ICheckAndRepairResults."""
2105 class IDeepCheckable(Interface):
2106 def start_deep_check(verify=False, add_lease=False):
2107 """Check upon the health of me and everything I can reach.
2109 This is a recursive form of check(), useable only on dirnodes.
2111 I return a Monitor, with results that are an IDeepCheckResults
2114 TODO: If any of the directories I traverse are unrecoverable, the
2115 Monitor will report failure. If any of the files I check upon are
2116 unrecoverable, those problems will be reported in the
2117 IDeepCheckResults as usual, and the Monitor will not report a
2121 def start_deep_check_and_repair(verify=False, add_lease=False):
2122 """Check upon the health of me and everything I can reach. Repair
2123 anything that isn't healthy.
2125 This is a recursive form of check_and_repair(), useable only on
2128 I return a Monitor, with results that are an
2129 IDeepCheckAndRepairResults object.
2131 TODO: If any of the directories I traverse are unrecoverable, the
2132 Monitor will report failure. If any of the files I check upon are
2133 unrecoverable, those problems will be reported in the
2134 IDeepCheckResults as usual, and the Monitor will not report a
2139 class ICheckResults(Interface):
2140 """I contain the detailed results of a check/verify operation.
2143 def get_storage_index():
2144 """Return a string with the (binary) storage index."""
2146 def get_storage_index_string():
2147 """Return a string with the (printable) abbreviated storage index."""
2150 """Return the (string) URI of the object that was checked."""
2153 """Return a boolean, True if the file/dir is fully healthy, False if
2154 it is damaged in any way. Non-distributed LIT files always return
2157 def is_recoverable():
2158 """Return a boolean, True if the file/dir can be recovered, False if
2159 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2160 files always return True."""
2162 # the following methods all return None for non-distributed LIT files
2164 def get_happiness():
2165 """Return the happiness count of the file."""
2167 def get_encoding_needed():
2168 """Return 'k', the number of shares required for recovery."""
2170 def get_encoding_expected():
2171 """Return 'N', the number of total shares generated."""
2173 def get_share_counter_good():
2174 """Return the number of distinct good shares that were found. For
2175 mutable files, this counts shares for the 'best' version."""
2177 def get_share_counter_wrong():
2178 """For mutable files, return the number of shares for versions other
2179 than the 'best' one (which is defined as being the recoverable
2180 version with the highest sequence number, then the highest roothash).
2181 These are either leftover shares from an older version (perhaps on a
2182 server that was offline when an update occurred), shares from an
2183 unrecoverable newer version, or shares from an alternate current
2184 version that results from an uncoordinated write collision. For a
2185 healthy file, this will equal 0. For immutable files, this will
2188 def get_corrupt_shares():
2189 """Return a list of 'share locators', one for each share that was
2190 found to be corrupt (integrity failure). Each share locator is a list
2191 of (IServer, storage_index, sharenum)."""
2193 def get_incompatible_shares():
2194 """Return a list of 'share locators', one for each share that was
2195 found to be of an unknown format. Each share locator is a list of
2196 (IServer, storage_index, sharenum)."""
2198 def get_servers_responding():
2199 """Return a list of IServer objects, one for each server that
2200 responded to the share query (even if they said they didn't have
2201 shares, and even if they said they did have shares but then didn't
2202 send them when asked, or dropped the connection, or returned a
2203 Failure, and even if they said they did have shares and sent
2204 incorrect ones when asked)"""
2206 def get_host_counter_good_shares():
2207 """Return the number of distinct storage servers with good shares. If
2208 this number is less than get_share_counters()[good], then some shares
2209 are doubled up, increasing the correlation of failures. This
2210 indicates that one or more shares should be moved to an otherwise
2211 unused server, if one is available.
2214 def get_version_counter_recoverable():
2215 """Return the number of recoverable versions of the file. For a
2216 healthy file, this will equal 1."""
2218 def get_version_counter_unrecoverable():
2219 """Return the number of unrecoverable versions of the file. For a
2220 healthy file, this will be 0."""
2223 """Return a dict mapping share identifier to list of IServer objects.
2224 This indicates which servers are holding which shares. For immutable
2225 files, the shareid is an integer (the share number, from 0 to N-1).
2226 For mutable files, it is a string of the form 'seq%d-%s-sh%d',
2227 containing the sequence number, the roothash, and the share number."""
2230 """Return a string with a brief (one-line) summary of the results."""
2233 """Return a list of strings with more detailed results."""
2236 class ICheckAndRepairResults(Interface):
2237 """I contain the detailed results of a check/verify/repair operation.
2239 The IFilesystemNode.check()/verify()/repair() methods all return
2240 instances that provide ICheckAndRepairResults.
2243 def get_storage_index():
2244 """Return a string with the (binary) storage index."""
2246 def get_storage_index_string():
2247 """Return a string with the (printable) abbreviated storage index."""
2249 def get_repair_attempted():
2250 """Return a boolean, True if a repair was attempted. We might not
2251 attempt to repair the file because it was healthy, or healthy enough
2252 (i.e. some shares were missing but not enough to exceed some
2253 threshold), or because we don't know how to repair this object."""
2255 def get_repair_successful():
2256 """Return a boolean, True if repair was attempted and the file/dir
2257 was fully healthy afterwards. False if no repair was attempted or if
2258 a repair attempt failed."""
2260 def get_pre_repair_results():
2261 """Return an ICheckResults instance that describes the state of the
2262 file/dir before any repair was attempted."""
2264 def get_post_repair_results():
2265 """Return an ICheckResults instance that describes the state of the
2266 file/dir after any repair was attempted. If no repair was attempted,
2267 the pre-repair and post-repair results will be identical."""
2270 class IDeepCheckResults(Interface):
2271 """I contain the results of a deep-check operation.
2273 This is returned by a call to ICheckable.deep_check().
2276 def get_root_storage_index_string():
2277 """Return the storage index (abbreviated human-readable string) of
2278 the first object checked."""
2281 """Return a dictionary with the following keys::
2283 count-objects-checked: count of how many objects were checked
2284 count-objects-healthy: how many of those objects were completely
2286 count-objects-unhealthy: how many were damaged in some way
2287 count-objects-unrecoverable: how many were unrecoverable
2288 count-corrupt-shares: how many shares were found to have
2289 corruption, summed over all objects
2293 def get_corrupt_shares():
2294 """Return a set of (IServer, storage_index, sharenum) for all shares
2295 that were found to be corrupt. storage_index is binary."""
2297 def get_all_results():
2298 """Return a dictionary mapping pathname (a tuple of strings, ready to
2299 be slash-joined) to an ICheckResults instance, one for each object
2300 that was checked."""
2302 def get_results_for_storage_index(storage_index):
2303 """Retrive the ICheckResults instance for the given (binary)
2304 storage index. Raises KeyError if there are no results for that
2308 """Return a dictionary with the same keys as
2309 IDirectoryNode.deep_stats()."""
2312 class IDeepCheckAndRepairResults(Interface):
2313 """I contain the results of a deep-check-and-repair operation.
2315 This is returned by a call to ICheckable.deep_check_and_repair().
2318 def get_root_storage_index_string():
2319 """Return the storage index (abbreviated human-readable string) of
2320 the first object checked."""
2323 """Return a dictionary with the following keys::
2325 count-objects-checked: count of how many objects were checked
2326 count-objects-healthy-pre-repair: how many of those objects were
2327 completely healthy (before any
2329 count-objects-unhealthy-pre-repair: how many were damaged in
2331 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2332 count-objects-healthy-post-repair: how many of those objects were
2333 completely healthy (after any
2335 count-objects-unhealthy-post-repair: how many were damaged in
2337 count-objects-unrecoverable-post-repair: how many were
2339 count-repairs-attempted: repairs were attempted on this many
2340 objects. The count-repairs- keys will
2341 always be provided, however unless
2342 repair=true is present, they will all
2344 count-repairs-successful: how many repairs resulted in healthy
2346 count-repairs-unsuccessful: how many repairs resulted did not
2347 results in completely healthy objects
2348 count-corrupt-shares-pre-repair: how many shares were found to
2349 have corruption, summed over all
2350 objects examined (before any
2352 count-corrupt-shares-post-repair: how many shares were found to
2353 have corruption, summed over all
2354 objects examined (after any
2359 """Return a dictionary with the same keys as
2360 IDirectoryNode.deep_stats()."""
2362 def get_corrupt_shares():
2363 """Return a set of (IServer, storage_index, sharenum) for all shares
2364 that were found to be corrupt before any repair was attempted.
2365 storage_index is binary.
2367 def get_remaining_corrupt_shares():
2368 """Return a set of (IServer, storage_index, sharenum) for all shares
2369 that were found to be corrupt after any repair was completed.
2370 storage_index is binary. These are shares that need manual inspection
2371 and probably deletion.
2373 def get_all_results():
2374 """Return a dictionary mapping pathname (a tuple of strings, ready to
2375 be slash-joined) to an ICheckAndRepairResults instance, one for each
2376 object that was checked."""
2378 def get_results_for_storage_index(storage_index):
2379 """Retrive the ICheckAndRepairResults instance for the given (binary)
2380 storage index. Raises KeyError if there are no results for that
2384 class IRepairable(Interface):
2385 def repair(check_results):
2386 """Attempt to repair the given object. Returns a Deferred that fires
2387 with a IRepairResults object.
2389 I must be called with an object that implements ICheckResults, as
2390 proof that you have actually discovered a problem with this file. I
2391 will use the data in the checker results to guide the repair process,
2392 such as which servers provided bad data and should therefore be
2393 avoided. The ICheckResults object is inside the
2394 ICheckAndRepairResults object, which is returned by the
2395 ICheckable.check() method::
2397 d = filenode.check(repair=False)
2398 def _got_results(check_and_repair_results):
2399 check_results = check_and_repair_results.get_pre_repair_results()
2400 return filenode.repair(check_results)
2401 d.addCallback(_got_results)
2406 class IRepairResults(Interface):
2407 """I contain the results of a repair operation."""
2408 def get_successful():
2409 """Returns a boolean: True if the repair made the file healthy, False
2410 if not. Repair failure generally indicates a file that has been
2411 damaged beyond repair."""
2414 class IClient(Interface):
2415 def upload(uploadable):
2416 """Upload some data into a CHK, get back the UploadResults for it.
2417 @param uploadable: something that implements IUploadable
2418 @return: a Deferred that fires with the UploadResults instance.
2419 To get the URI for this file, use results.uri .
2422 def create_mutable_file(contents=""):
2423 """Create a new mutable file (with initial) contents, get back the
2426 @param contents: (bytestring, callable, or None): this provides the
2427 initial contents of the mutable file. If 'contents' is a bytestring,
2428 it will be used as-is. If 'contents' is a callable, it will be
2429 invoked with the new MutableFileNode instance and is expected to
2430 return a bytestring with the initial contents of the file (the
2431 callable can use node.get_writekey() to decide how to encrypt the
2432 initial contents, e.g. for a brand new dirnode with initial
2433 children). contents=None is equivalent to an empty string. Using
2434 content_maker= is more efficient than creating a mutable file and
2435 setting its contents in two separate operations.
2437 @return: a Deferred that fires with an IMutableFileNode instance.
2440 def create_dirnode(initial_children={}):
2441 """Create a new unattached dirnode, possibly with initial children.
2443 @param initial_children: dict with keys that are unicode child names,
2444 and values that are (childnode, metadata) tuples.
2446 @return: a Deferred that fires with the new IDirectoryNode instance.
2449 def create_node_from_uri(uri, rouri):
2450 """Create a new IFilesystemNode instance from the uri, synchronously.
2451 @param uri: a string or IURI-providing instance, or None. This could
2452 be for a LiteralFileNode, a CHK file node, a mutable file
2453 node, or a directory node
2454 @param rouri: a string or IURI-providing instance, or None. If the
2455 main uri is None, I will use the rouri instead. If I
2456 recognize the format of the main uri, I will ignore the
2457 rouri (because it can be derived from the writecap).
2459 @return: an instance that provides IFilesystemNode (or more usefully
2460 one of its subclasses). File-specifying URIs will result in
2461 IFileNode-providing instances, like ImmutableFileNode,
2462 LiteralFileNode, or MutableFileNode. Directory-specifying
2463 URIs will result in IDirectoryNode-providing instances, like
2468 class INodeMaker(Interface):
2469 """The NodeMaker is used to create IFilesystemNode instances. It can
2470 accept a filecap/dircap string and return the node right away. It can
2471 also create new nodes (i.e. upload a file, or create a mutable file)
2472 asynchronously. Once you have one of these nodes, you can use other
2473 methods to determine whether it is a file or directory, and to download
2474 or modify its contents.
2476 The NodeMaker encapsulates all the authorities that these
2477 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2478 Tahoe process will typically have a single NodeMaker, but unit tests may
2479 create simplified/mocked forms for testing purposes.
2482 def create_from_cap(writecap, readcap=None, deep_immutable=False, name=u"<unknown name>"):
2483 """I create an IFilesystemNode from the given writecap/readcap. I can
2484 only provide nodes for existing file/directory objects: use my other
2485 methods to create new objects. I return synchronously."""
2487 def create_mutable_file(contents=None, keysize=None):
2488 """I create a new mutable file, and return a Deferred that will fire
2489 with the IMutableFileNode instance when it is ready. If contents= is
2490 provided (a bytestring), it will be used as the initial contents of
2491 the new file, otherwise the file will contain zero bytes. keysize= is
2492 for use by unit tests, to create mutable files that are smaller than
2495 def create_new_mutable_directory(initial_children={}):
2496 """I create a new mutable directory, and return a Deferred that will
2497 fire with the IDirectoryNode instance when it is ready. If
2498 initial_children= is provided (a dict mapping unicode child name to
2499 (childnode, metadata_dict) tuples), the directory will be populated
2500 with those children, otherwise it will be empty."""
2503 class IClientStatus(Interface):
2504 def list_all_uploads():
2505 """Return a list of uploader objects, one for each upload that
2506 currently has an object available (tracked with weakrefs). This is
2507 intended for debugging purposes."""
2509 def list_active_uploads():
2510 """Return a list of active IUploadStatus objects."""
2512 def list_recent_uploads():
2513 """Return a list of IUploadStatus objects for the most recently
2516 def list_all_downloads():
2517 """Return a list of downloader objects, one for each download that
2518 currently has an object available (tracked with weakrefs). This is
2519 intended for debugging purposes."""
2521 def list_active_downloads():
2522 """Return a list of active IDownloadStatus objects."""
2524 def list_recent_downloads():
2525 """Return a list of IDownloadStatus objects for the most recently
2526 started downloads."""
2529 class IUploadStatus(Interface):
2531 """Return a timestamp (float with seconds since epoch) indicating
2532 when the operation was started."""
2534 def get_storage_index():
2535 """Return a string with the (binary) storage index in use on this
2536 upload. Returns None if the storage index has not yet been
2540 """Return an integer with the number of bytes that will eventually
2541 be uploaded for this file. Returns None if the size is not yet known.
2544 """Return True if this upload is using a Helper, False if not."""
2547 """Return a string describing the current state of the upload
2551 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2552 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2553 made towards hashing the file to determine a CHK encryption key: if
2554 non-convergent encryption is in use, this will be trivial, otherwise
2555 the whole file must be hashed. 'ciphertext' describes how much of the
2556 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2557 uploads. 'encode_and_push' describes how much of the encode-and-push
2558 process has finished: for helper uploads this is dependent upon the
2559 helper providing progress reports. It might be reasonable to add all
2560 three numbers and report the sum to the user."""
2563 """Return True if the upload is currently active, False if not."""
2566 """Return an instance of UploadResults (which contains timing and
2567 sharemap information). Might return None if the upload is not yet
2571 """Each upload status gets a unique number: this method returns that
2572 number. This provides a handle to this particular upload, so a web
2573 page can generate a suitable hyperlink."""
2576 class IDownloadStatus(Interface):
2578 """Return a timestamp (float with seconds since epoch) indicating
2579 when the operation was started."""
2581 def get_storage_index():
2582 """Return a string with the (binary) storage index in use on this
2583 download. This may be None if there is no storage index (i.e. LIT
2587 """Return an integer with the number of bytes that will eventually be
2588 retrieved for this file. Returns None if the size is not yet known.
2592 """Return True if this download is using a Helper, False if not."""
2595 """Return a string describing the current state of the download
2599 """Returns a float (from 0.0 to 1.0) describing the amount of the
2600 download that has completed. This value will remain at 0.0 until the
2601 first byte of plaintext is pushed to the download target."""
2604 """Return True if the download is currently active, False if not."""
2607 """Each download status gets a unique number: this method returns
2608 that number. This provides a handle to this particular download, so a
2609 web page can generate a suitable hyperlink."""
2612 class IServermapUpdaterStatus(Interface):
2615 class IPublishStatus(Interface):
2618 class IRetrieveStatus(Interface):
2622 class NotCapableError(Exception):
2623 """You have tried to write to a read-only node."""
2625 class BadWriteEnablerError(Exception):
2629 class RIControlClient(RemoteInterface):
2630 def wait_for_client_connections(num_clients=int):
2631 """Do not return until we have connections to at least NUM_CLIENTS
2637 def upload_random_data_from_file(size=int, convergence=str):
2640 def download_to_tempfile_and_delete(uri=str):
2643 def get_memory_usage():
2644 """Return a dict describes the amount of memory currently in use. The
2645 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2646 measuring memory consupmtion in bytes."""
2647 return DictOf(str, int)
2649 def speed_test(count=int, size=int, mutable=Any()):
2650 """Write 'count' tempfiles to disk, all of the given size. Measure
2651 how long (in seconds) it takes to upload them all to the servers.
2652 Then measure how long it takes to download all of them. If 'mutable'
2653 is 'create', time creation of mutable files. If 'mutable' is
2654 'upload', then time access to the same mutable file instead of
2657 Returns a tuple of (upload_time, download_time).
2659 return (float, float)
2661 def measure_peer_response_time():
2662 """Send a short message to each connected peer, and measure the time
2663 it takes for them to respond to it. This is a rough measure of the
2664 application-level round trip time.
2666 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2669 return DictOf(str, float)
2672 UploadResults = Any() #DictOf(str, str)
2675 class RIEncryptedUploadable(RemoteInterface):
2676 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2681 def get_all_encoding_parameters():
2682 return (int, int, int, long)
2684 def read_encrypted(offset=Offset, length=ReadSize):
2691 class RICHKUploadHelper(RemoteInterface):
2692 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2696 Return a dictionary of version information.
2698 return DictOf(str, Any())
2700 def upload(reader=RIEncryptedUploadable):
2701 return UploadResults
2704 class RIHelper(RemoteInterface):
2705 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2709 Return a dictionary of version information.
2711 return DictOf(str, Any())
2713 def upload_chk(si=StorageIndex):
2714 """See if a file with a given storage index needs uploading. The
2715 helper will ask the appropriate storage servers to see if the file
2716 has already been uploaded. If so, the helper will return a set of
2717 'upload results' that includes whatever hashes are needed to build
2718 the read-cap, and perhaps a truncated sharemap.
2720 If the file has not yet been uploaded (or if it was only partially
2721 uploaded), the helper will return an empty upload-results dictionary
2722 and also an RICHKUploadHelper object that will take care of the
2723 upload process. The client should call upload() on this object and
2724 pass it a reference to an RIEncryptedUploadable object that will
2725 provide ciphertext. When the upload is finished, the upload() method
2726 will finish and return the upload results.
2728 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2731 class RIStatsProvider(RemoteInterface):
2732 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2734 Provides access to statistics and monitoring information.
2739 returns a dictionary containing 'counters' and 'stats', each a
2740 dictionary with string counter/stat name keys, and numeric or None values.
2741 counters are monotonically increasing measures of work done, and
2742 stats are instantaneous measures (potentially time averaged
2745 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2748 class RIStatsGatherer(RemoteInterface):
2749 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2751 Provides a monitoring service for centralised collection of stats
2754 def provide(provider=RIStatsProvider, nickname=str):
2756 @param provider: a stats collector instance that should be polled
2757 periodically by the gatherer to collect stats.
2758 @param nickname: a name useful to identify the provided client
2763 class IStatsProducer(Interface):
2766 returns a dictionary, with str keys representing the names of stats
2767 to be monitored, and numeric values.
2770 class RIKeyGenerator(RemoteInterface):
2771 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2773 Provides a service offering to make RSA key pairs.
2776 def get_rsa_key_pair(key_size=int):
2778 @param key_size: the size of the signature key.
2779 @return: tuple(verifying_key, signing_key)
2781 return TupleOf(str, str)
2784 class FileTooLargeError(Exception):
2788 class IValidatedThingProxy(Interface):
2790 """ Acquire a thing and validate it. Return a deferred that is
2791 eventually fired with self if the thing is valid or errbacked if it
2792 can't be acquired or validated."""
2795 class InsufficientVersionError(Exception):
2796 def __init__(self, needed, got):
2797 self.needed = needed
2801 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2804 class EmptyPathnameComponentError(Exception):
2805 """The webapi disallows empty pathname components."""