2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
8 Hash = StringConstraint(maxLength=HASH_SIZE,
9 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
10 Nodeid = StringConstraint(maxLength=20,
11 minLength=20) # binary format 20-byte SHA1 hash
12 FURL = StringConstraint(1000)
13 StorageIndex = StringConstraint(16)
14 URI = StringConstraint(300) # kind of arbitrary
16 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
18 ShareData = StringConstraint(None)
19 URIExtensionData = StringConstraint(1000)
20 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
22 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
23 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
24 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
25 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
27 class RIStubClient(RemoteInterface):
28 """Each client publishes a service announcement for a dummy object called
29 the StubClient. This object doesn't actually offer any services, but the
30 announcement helps the Introducer keep track of which clients are
31 subscribed (so the grid admin can keep track of things like the size of
32 the grid and the client versions in use. This is the (empty)
33 RemoteInterface for the StubClient."""
35 class RIBucketWriter(RemoteInterface):
36 """ Objects of this kind live on the server side. """
37 def write(offset=Offset, data=ShareData):
42 If the data that has been written is incomplete or inconsistent then
43 the server will throw the data away, else it will store it for future
49 """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 which is tied to a specific share, and therefore does not need the
65 extra share-identifying arguments. Please see that method for full
69 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
70 # elements are (offset, length, operator, specimen)
71 # operator is one of "lt, le, eq, ne, ge, gt"
72 # nop always passes and is used to fetch data while writing.
73 # you should use length==len(specimen) for everything except nop
74 DataVector = ListOf(TupleOf(Offset, ShareData))
75 # (offset, data). This limits us to 30 writes of 1MiB each per call
76 TestAndWriteVectorsForShares = DictOf(int,
79 ChoiceOf(None, Offset), # new_length
81 ReadVector = ListOf(TupleOf(Offset, ReadSize))
82 ReadData = ListOf(ShareData)
83 # returns data[offset:offset+length] for each element of TestVector
85 class RIStorageServer(RemoteInterface):
86 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
90 Return a dictionary of version information.
92 return DictOf(str, Any())
94 def allocate_buckets(storage_index=StorageIndex,
95 renew_secret=LeaseRenewSecret,
96 cancel_secret=LeaseCancelSecret,
97 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
98 allocated_size=Offset, canary=Referenceable):
100 @param storage_index: the index of the bucket to be created or
102 @param sharenums: these are the share numbers (probably between 0 and
103 99) that the sender is proposing to store on this
105 @param renew_secret: This is the secret used to protect bucket refresh
106 This secret is generated by the client and
107 stored for later comparison by the server. Each
108 server is given a different secret.
109 @param cancel_secret: Like renew_secret, but protects bucket decref.
110 @param canary: If the canary is lost before close(), the bucket is
112 @return: tuple of (alreadygot, allocated), where alreadygot is what we
113 already have and allocated is what we hereby agree to accept.
114 New leases are added for shares in both lists.
116 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
117 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
119 def add_lease(storage_index=StorageIndex,
120 renew_secret=LeaseRenewSecret,
121 cancel_secret=LeaseCancelSecret):
123 Add a new lease on the given bucket. If the renew_secret matches an
124 existing lease, that lease will be renewed instead. If there is no
125 bucket for the given storage_index, return silently. (note that in
126 tahoe-1.3.0 and earlier, IndexError was raised if there was no
129 return Any() # returns None now, but future versions might change
131 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
133 Renew the lease on a given bucket, resetting the timer to 31 days.
134 Some networks will use this, some will not. If there is no bucket for
135 the given storage_index, IndexError will be raised.
137 For mutable shares, if the given renew_secret does not match an
138 existing lease, IndexError will be raised with a note listing the
139 server-nodeids on the existing leases, so leases on migrated shares
140 can be renewed or cancelled. For immutable shares, IndexError
141 (without the note) will be raised.
145 def cancel_lease(storage_index=StorageIndex,
146 cancel_secret=LeaseCancelSecret):
148 Cancel the lease on a given bucket. If this was the last lease on the
149 bucket, the bucket will be deleted. If there is no bucket for the
150 given storage_index, IndexError will be raised.
152 For mutable shares, if the given cancel_secret does not match an
153 existing lease, IndexError will be raised with a note listing the
154 server-nodeids on the existing leases, so leases on migrated shares
155 can be renewed or cancelled. For immutable shares, IndexError
156 (without the note) will be raised.
160 def get_buckets(storage_index=StorageIndex):
161 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
165 def slot_readv(storage_index=StorageIndex,
166 shares=ListOf(int), readv=ReadVector):
167 """Read a vector from the numbered shares associated with the given
168 storage index. An empty shares list means to return data from all
169 known shares. Returns a dictionary with one key per share."""
170 return DictOf(int, ReadData) # shnum -> results
172 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
173 secrets=TupleOf(WriteEnablerSecret,
176 tw_vectors=TestAndWriteVectorsForShares,
179 """General-purpose test-and-set operation for mutable slots. Perform
180 a bunch of comparisons against the existing shares. If they all pass,
181 then apply a bunch of write vectors to those shares. Then use the
182 read vectors to extract data from all the shares and return the data.
184 This method is, um, large. The goal is to allow clients to update all
185 the shares associated with a mutable file in a single round trip.
187 @param storage_index: the index of the bucket to be created or
189 @param write_enabler: a secret that is stored along with the slot.
190 Writes are accepted from any caller who can
191 present the matching secret. A different secret
192 should be used for each slot*server pair.
193 @param renew_secret: This is the secret used to protect bucket refresh
194 This secret is generated by the client and
195 stored for later comparison by the server. Each
196 server is given a different secret.
197 @param cancel_secret: Like renew_secret, but protects bucket decref.
199 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
200 cancel_secret). The first is required to perform any write. The
201 latter two are used when allocating new shares. To simply acquire a
202 new lease on existing shares, use an empty testv and an empty writev.
204 Each share can have a separate test vector (i.e. a list of
205 comparisons to perform). If all vectors for all shares pass, then all
206 writes for all shares are recorded. Each comparison is a 4-tuple of
207 (offset, length, operator, specimen), which effectively does a bool(
208 (read(offset, length)) OPERATOR specimen ) and only performs the
209 write if all these evaluate to True. Basic test-and-set uses 'eq'.
210 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
211 Write-if-same-or-newer uses 'le'.
213 Reads from the end of the container are truncated, and missing shares
214 behave like empty ones, so to assert that a share doesn't exist (for
215 use when creating a new share), use (0, 1, 'eq', '').
217 The write vector will be applied to the given share, expanding it if
218 necessary. A write vector applied to a share number that did not
219 exist previously will cause that share to be created.
221 Each write vector is accompanied by a 'new_length' argument. If
222 new_length is not None, use it to set the size of the container. This
223 can be used to pre-allocate space for a series of upcoming writes, or
224 truncate existing data. If the container is growing, new_length will
225 be applied before datav. If the container is shrinking, it will be
226 applied afterwards. If new_length==0, the share will be deleted.
228 The read vector is used to extract data from all known shares,
229 *before* any writes have been applied. The same vector is used for
230 all shares. This captures the state that was tested by the test
233 This method returns two values: a boolean and a dict. The boolean is
234 True if the write vectors were applied, False if not. The dict is
235 keyed by share number, and each value contains a list of strings, one
236 for each element of the read vector.
238 If the write_enabler is wrong, this will raise BadWriteEnablerError.
239 To enable share migration (using update_write_enabler), the exception
240 will have the nodeid used for the old write enabler embedded in it,
241 in the following string::
243 The write enabler was recorded by nodeid '%s'.
245 Note that the nodeid here is encoded using the same base32 encoding
246 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
249 return TupleOf(bool, DictOf(int, ReadData))
251 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
252 shnum=int, reason=str):
253 """Clients who discover hash failures in shares that they have
254 downloaded from me will use this method to inform me about the
255 failures. I will record their concern so that my operator can
256 manually inspect the shares in question. I return None.
258 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
259 (binary) storage index string, and 'shnum' is the integer share
260 number. 'reason' is a human-readable explanation of the problem,
261 probably including some expected hash values and the computed ones
262 which did not match. Corruption advisories for mutable shares should
263 include a hash of the public key (the same value that appears in the
264 mutable-file verify-cap), since the current share format does not
268 class IStorageBucketWriter(Interface):
270 Objects of this kind live on the client side.
272 def put_block(segmentnum=int, data=ShareData):
273 """@param data: For most segments, this data will be 'blocksize'
274 bytes in length. The last segment might be shorter.
275 @return: a Deferred that fires (with None) when the operation completes
278 def put_plaintext_hashes(hashes=ListOf(Hash)):
280 @return: a Deferred that fires (with None) when the operation completes
283 def put_crypttext_hashes(hashes=ListOf(Hash)):
285 @return: a Deferred that fires (with None) when the operation completes
288 def put_block_hashes(blockhashes=ListOf(Hash)):
290 @return: a Deferred that fires (with None) when the operation completes
293 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
295 @return: a Deferred that fires (with None) when the operation completes
298 def put_uri_extension(data=URIExtensionData):
299 """This block of data contains integrity-checking information (hashes
300 of plaintext, crypttext, and shares), as well as encoding parameters
301 that are necessary to recover the data. This is a serialized dict
302 mapping strings to other strings. The hash of this data is kept in
303 the URI and verified before any of the data is used. All buckets for
304 a given file contain identical copies of this data.
306 The serialization format is specified with the following pseudocode:
307 for k in sorted(dict.keys()):
308 assert re.match(r'^[a-zA-Z_\-]+$', k)
309 write(k + ':' + netstring(dict[k]))
311 @return: a Deferred that fires (with None) when the operation completes
315 """Finish writing and close the bucket. The share is not finalized
316 until this method is called: if the uploading client disconnects
317 before calling close(), the partially-written share will be
320 @return: a Deferred that fires (with None) when the operation completes
323 class IStorageBucketReader(Interface):
325 def get_block_data(blocknum=int, blocksize=int, size=int):
326 """Most blocks will be the same size. The last block might be shorter
332 def get_crypttext_hashes():
334 @return: ListOf(Hash)
337 def get_block_hashes(at_least_these=SetOf(int)):
339 @return: ListOf(Hash)
342 def get_share_hashes(at_least_these=SetOf(int)):
344 @return: ListOf(TupleOf(int, Hash))
347 def get_uri_extension():
349 @return: URIExtensionData
352 class IStorageBroker(Interface):
353 def get_servers_for_index(peer_selection_index):
355 @return: list of (peerid, versioned-rref) tuples
357 def get_all_servers():
359 @return: frozenset of (peerid, versioned-rref) tuples
361 def get_all_serverids():
363 @return: frozenset of serverid strings
365 def get_nickname_for_serverid(serverid):
367 @return: unicode nickname, or None
370 # methods moved from IntroducerClient, need review
371 def get_all_connections():
372 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
373 each active connection we've established to a remote service. This is
374 mostly useful for unit tests that need to wait until a certain number
375 of connections have been made."""
377 def get_all_connectors():
378 """Return a dict that maps from (nodeid, service_name) to a
379 RemoteServiceConnector instance for all services that we are actively
380 trying to connect to. Each RemoteServiceConnector has the following
383 service_name: the type of service provided, like 'storage'
384 announcement_time: when we first heard about this service
385 last_connect_time: when we last established a connection
386 last_loss_time: when we last lost a connection
388 version: the peer's version, from the most recent connection
389 oldest_supported: the peer's oldest supported version, same
391 rref: the RemoteReference, if connected, otherwise None
392 remote_host: the IAddress, if connected, otherwise None
394 This method is intended for monitoring interfaces, such as a web page
395 which describes connecting and connected peers.
398 def get_all_peerids():
399 """Return a frozenset of all peerids to whom we have a connection (to
400 one or more services) established. Mostly useful for unit tests."""
402 def get_all_connections_for(service_name):
403 """Return a frozenset of (nodeid, service_name, rref) tuples, one
404 for each active connection that provides the given SERVICE_NAME."""
406 def get_permuted_peers(service_name, key):
407 """Returns an ordered list of (peerid, rref) tuples, selecting from
408 the connections that provide SERVICE_NAME, using a hash-based
409 permutation keyed by KEY. This randomizes the service list in a
410 repeatable way, to distribute load over many peers.
414 class IURI(Interface):
415 def init_from_string(uri):
416 """Accept a string (as created by my to_string() method) and populate
417 this instance with its data. I am not normally called directly,
418 please use the module-level uri.from_string() function to convert
419 arbitrary URI strings into IURI-providing instances."""
422 """Return False if this URI be used to modify the data. Return True
423 if this URI cannot be used to modify the data."""
426 """Return True if the data can be modified by *somebody* (perhaps
427 someone who has a more powerful URI than this one)."""
430 """Return another IURI instance, which represents a read-only form of
431 this one. If is_readonly() is True, this returns self."""
433 def get_verify_cap():
434 """Return an instance that provides IVerifierURI, which can be used
435 to check on the availability of the file or directory, without
436 providing enough capabilities to actually read or modify the
437 contents. This may return None if the file does not need checking or
438 verification (e.g. LIT URIs).
442 """Return a string of printable ASCII characters, suitable for
443 passing into init_from_string."""
445 class IVerifierURI(Interface, IURI):
446 def init_from_string(uri):
447 """Accept a string (as created by my to_string() method) and populate
448 this instance with its data. I am not normally called directly,
449 please use the module-level uri.from_string() function to convert
450 arbitrary URI strings into IURI-providing instances."""
453 """Return a string of printable ASCII characters, suitable for
454 passing into init_from_string."""
456 class IDirnodeURI(Interface):
457 """I am a URI which represents a dirnode."""
459 class IFileURI(Interface):
460 """I am a URI which represents a filenode."""
462 """Return the length (in bytes) of the file that I represent."""
464 class IImmutableFileURI(IFileURI):
467 class IMutableFileURI(Interface):
468 """I am a URI which represents a mutable filenode."""
470 class IDirectoryURI(Interface):
473 class IReadonlyDirectoryURI(Interface):
476 class CapConstraintError(Exception):
477 """A constraint on a cap was violated."""
479 class MustBeDeepImmutableError(CapConstraintError):
480 """Mutable children cannot be added to an immutable directory.
481 Also, caps obtained from an immutable directory can trigger this error
482 if they are later found to refer to a mutable object and then used."""
484 class MustBeReadonlyError(CapConstraintError):
485 """Known write caps cannot be specified in a ro_uri field. Also,
486 caps obtained from a ro_uri field can trigger this error if they
487 are later found to be write caps and then used."""
489 class MustNotBeUnknownRWError(CapConstraintError):
490 """Cannot add an unknown child cap specified in a rw_uri field."""
492 # The hierarchy looks like this:
499 class IFilesystemNode(Interface):
501 """Return the strongest 'cap instance' associated with this node.
502 (writecap for writeable-mutable files/directories, readcap for
503 immutable or readonly-mutable files/directories). To convert this
504 into a string, call .to_string() on the result."""
507 """Return a readonly cap instance for this node. For immutable or
508 readonly nodes, get_cap() and get_readcap() return the same thing."""
510 def get_repair_cap():
511 """Return an IURI instance that can be used to repair the file, or
512 None if this node cannot be repaired (either because it is not
513 distributed, like a LIT file, or because the node does not represent
514 sufficient authority to create a repair-cap, like a read-only RSA
515 mutable file node [which cannot create the correct write-enablers]).
518 def get_verify_cap():
519 """Return an IVerifierURI instance that represents the
520 'verifiy/refresh capability' for this node. The holder of this
521 capability will be able to renew the lease for this node, protecting
522 it from garbage-collection. They will also be able to ask a server if
523 it holds a share for the file or directory.
527 """Return the URI string corresponding to the strongest cap associated
528 with this node. If this node is read-only, the URI will only offer
529 read-only access. If this node is read-write, the URI will offer
532 If you have read-write access to a node and wish to share merely
533 read-only access with others, use get_readonly_uri().
536 def get_write_uri(n):
537 """Return the URI string that can be used by others to get write
538 access to this node, if it is writeable. If this is a read-only node,
541 def get_readonly_uri():
542 """Return the URI string that can be used by others to get read-only
543 access to this node. The result is a read-only URI, regardless of
544 whether this node is read-only or read-write.
546 If you have merely read-only access to this node, get_readonly_uri()
547 will return the same thing as get_uri().
550 def get_storage_index():
551 """Return a string with the (binary) storage index in use on this
552 download. This may be None if there is no storage index (i.e. LIT
556 """Return True if this reference provides mutable access to the given
557 file or directory (i.e. if you can modify it), or False if not. Note
558 that even if this reference is read-only, someone else may hold a
559 read-write reference to it."""
562 """Return True if this file or directory is mutable (by *somebody*,
563 not necessarily you), False if it is is immutable. Note that a file
564 might be mutable overall, but your reference to it might be
565 read-only. On the other hand, all references to an immutable file
566 will be read-only; there are no read-write references to an immutable
571 """Return True if this is an unknown node."""
573 def is_allowed_in_immutable_directory():
574 """Return True if this node is allowed as a child of a deep-immutable
575 directory. This is true if either the node is of a known-immutable type,
576 or it is unknown and read-only.
580 """Raise any error associated with this node."""
583 """Return the length (in bytes) of the data this node represents. For
584 directory nodes, I return the size of the backing store. I return
585 synchronously and do not consult the network, so for mutable objects,
586 I will return the most recently observed size for the object, or None
587 if I don't remember a size. Use get_current_size, which returns a
588 Deferred, if you want more up-to-date information."""
590 def get_current_size():
591 """I return a Deferred that fires with the length (in bytes) of the
592 data this node represents.
595 class IFileNode(IFilesystemNode):
596 """I am a node which represents a file: a sequence of bytes. I am not a
597 container, like IDirectoryNode."""
599 class IImmutableFileNode(IFileNode):
600 def read(consumer, offset=0, size=None):
601 """Download a portion (possibly all) of the file's contents, making
602 them available to the given IConsumer. Return a Deferred that fires
603 (with the consumer) when the consumer is unregistered (either because
604 the last byte has been given to it, or because the consumer threw an
605 exception during write(), possibly because it no longer wants to
606 receive data). The portion downloaded will start at 'offset' and
607 contain 'size' bytes (or the remainder of the file if size==None).
609 The consumer will be used in non-streaming mode: an IPullProducer
610 will be attached to it.
612 The consumer will not receive data right away: several network trips
613 must occur first. The order of events will be::
615 consumer.registerProducer(p, streaming)
616 (if streaming == False)::
617 consumer does p.resumeProducing()
619 consumer does p.resumeProducing()
620 consumer.write(data).. (repeat until all data is written)
621 consumer.unregisterProducer()
622 deferred.callback(consumer)
624 If a download error occurs, or an exception is raised by
625 consumer.registerProducer() or consumer.write(), I will call
626 consumer.unregisterProducer() and then deliver the exception via
627 deferred.errback(). To cancel the download, the consumer should call
628 p.stopProducing(), which will result in an exception being delivered
629 via deferred.errback().
631 See src/allmydata/util/consumer.py for an example of a simple
632 download-to-memory consumer.
635 class IMutableFileNode(IFileNode):
636 """I provide access to a 'mutable file', which retains its identity
637 regardless of what contents are put in it.
639 The consistency-vs-availability problem means that there might be
640 multiple versions of a file present in the grid, some of which might be
641 unrecoverable (i.e. have fewer than 'k' shares). These versions are
642 loosely ordered: each has a sequence number and a hash, and any version
643 with seqnum=N was uploaded by a node which has seen at least one version
646 The 'servermap' (an instance of IMutableFileServerMap) is used to
647 describe the versions that are known to be present in the grid, and which
648 servers are hosting their shares. It is used to represent the 'state of
649 the world', and is used for this purpose by my test-and-set operations.
650 Downloading the contents of the mutable file will also return a
651 servermap. Uploading a new version into the mutable file requires a
652 servermap as input, and the semantics of the replace operation is
653 'replace the file with my new version if it looks like nobody else has
654 changed the file since my previous download'. Because the file is
655 distributed, this is not a perfect test-and-set operation, but it will do
656 its best. If the replace process sees evidence of a simultaneous write,
657 it will signal an UncoordinatedWriteError, so that the caller can take
661 Most readers will want to use the 'best' current version of the file, and
662 should use my 'download_best_version()' method.
664 To unconditionally replace the file, callers should use overwrite(). This
665 is the mode that user-visible mutable files will probably use.
667 To apply some delta to the file, call modify() with a callable modifier
668 function that can apply the modification that you want to make. This is
669 the mode that dirnodes will use, since most directory modification
670 operations can be expressed in terms of deltas to the directory state.
673 Three methods are available for users who need to perform more complex
674 operations. The first is get_servermap(), which returns an up-to-date
675 servermap using a specified mode. The second is download_version(), which
676 downloads a specific version (not necessarily the 'best' one). The third
677 is 'upload', which accepts new contents and a servermap (which must have
678 been updated with MODE_WRITE). The upload method will attempt to apply
679 the new contents as long as no other node has modified the file since the
680 servermap was updated. This might be useful to a caller who wants to
681 merge multiple versions into a single new one.
683 Note that each time the servermap is updated, a specific 'mode' is used,
684 which determines how many peers are queried. To use a servermap for my
685 replace() method, that servermap must have been updated in MODE_WRITE.
686 These modes are defined in allmydata.mutable.common, and consist of
687 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
688 allmydata/mutable/servermap.py for details about the differences.
690 Mutable files are currently limited in size (about 3.5MB max) and can
691 only be retrieved and updated all-at-once, as a single big string. Future
692 versions of our mutable files will remove this restriction.
695 def download_best_version():
696 """Download the 'best' available version of the file, meaning one of
697 the recoverable versions with the highest sequence number. If no
698 uncoordinated writes have occurred, and if enough shares are
699 available, then this will be the most recent version that has been
702 I update an internal servermap with MODE_READ, determine which
703 version of the file is indicated by
704 servermap.best_recoverable_version(), and return a Deferred that
705 fires with its contents. If no version is recoverable, the Deferred
706 will errback with UnrecoverableFileError.
709 def get_size_of_best_version():
710 """Find the size of the version that would be downloaded with
711 download_best_version(), without actually downloading the whole file.
713 I return a Deferred that fires with an integer.
716 def overwrite(new_contents):
717 """Unconditionally replace the contents of the mutable file with new
718 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
719 is only appropriate to use when the new contents of the file are
720 completely unrelated to the old ones, and you do not care about other
723 I return a Deferred that fires (with a PublishStatus object) when the
724 update has completed.
727 def modify(modifier_cb):
728 """Modify the contents of the file, by downloading the current
729 version, applying the modifier function (or bound method), then
730 uploading the new version. I return a Deferred that fires (with a
731 PublishStatus object) when the update is complete.
733 The modifier callable will be given three arguments: a string (with
734 the old contents), a 'first_time' boolean, and a servermap. As with
735 download_best_version(), the old contents will be from the best
736 recoverable version, but the modifier can use the servermap to make
737 other decisions (such as refusing to apply the delta if there are
738 multiple parallel versions, or if there is evidence of a newer
739 unrecoverable version). 'first_time' will be True the first time the
740 modifier is called, and False on any subsequent calls.
742 The callable should return a string with the new contents. The
743 callable must be prepared to be called multiple times, and must
744 examine the input string to see if the change that it wants to make
745 is already present in the old version. If it does not need to make
746 any changes, it can either return None, or return its input string.
748 If the modifier raises an exception, it will be returned in the
753 def get_servermap(mode):
754 """Return a Deferred that fires with an IMutableFileServerMap
755 instance, updated using the given mode.
758 def download_version(servermap, version):
759 """Download a specific version of the file, using the servermap
760 as a guide to where the shares are located.
762 I return a Deferred that fires with the requested contents, or
763 errbacks with UnrecoverableFileError. Note that a servermap which was
764 updated with MODE_ANYTHING or MODE_READ may not know about shares for
765 all versions (those modes stop querying servers as soon as they can
766 fulfil their goals), so you may want to use MODE_CHECK (which checks
767 everything) to get increased visibility.
770 def upload(new_contents, servermap):
771 """Replace the contents of the file with new ones. This requires a
772 servermap that was previously updated with MODE_WRITE.
774 I attempt to provide test-and-set semantics, in that I will avoid
775 modifying any share that is different than the version I saw in the
776 servermap. However, if another node is writing to the file at the
777 same time as me, I may manage to update some shares while they update
778 others. If I see any evidence of this, I will signal
779 UncoordinatedWriteError, and the file will be left in an inconsistent
780 state (possibly the version you provided, possibly the old version,
781 possibly somebody else's version, and possibly a mix of shares from
784 The recommended response to UncoordinatedWriteError is to either
785 return it to the caller (since they failed to coordinate their
786 writes), or to attempt some sort of recovery. It may be sufficient to
787 wait a random interval (with exponential backoff) and repeat your
788 operation. If I do not signal UncoordinatedWriteError, then I was
789 able to write the new version without incident.
791 I return a Deferred that fires (with a PublishStatus object) when the
792 publish has completed. I will update the servermap in-place with the
793 location of all new shares.
797 """Return this filenode's writekey, or None if the node does not have
798 write-capability. This may be used to assist with data structures
799 that need to make certain data available only to writers, such as the
800 read-write child caps in dirnodes. The recommended process is to have
801 reader-visible data be submitted to the filenode in the clear (where
802 it will be encrypted by the filenode using the readkey), but encrypt
803 writer-visible data using this writekey.
806 class NotEnoughSharesError(Exception):
807 """Download was unable to get enough shares, or upload was unable to
808 place 'shares_of_happiness' shares."""
810 class NoSharesError(Exception):
811 """Upload or Download was unable to get any shares at all."""
813 class UnableToFetchCriticalDownloadDataError(Exception):
814 """I was unable to fetch some piece of critical data which is supposed to
815 be identically present in all shares."""
817 class NoServersError(Exception):
818 """Upload wasn't given any servers to work with, usually indicating a
819 network or Introducer problem."""
821 class ExistingChildError(Exception):
822 """A directory node was asked to add or replace a child that already
823 exists, and overwrite= was set to False."""
825 class NoSuchChildError(Exception):
826 """A directory node was asked to fetch a child which does not exist."""
828 class IDirectoryNode(IFilesystemNode):
829 """I represent a filesystem node that is a container, with a
830 name-to-child mapping, holding the tahoe equivalent of a directory. All
831 child names are unicode strings, and all children are some sort of
832 IFilesystemNode (either files or subdirectories).
837 The dirnode ('1') URI returned by this method can be used in
838 set_uri() on a different directory ('2') to 'mount' a reference to
839 this directory ('1') under the other ('2'). This URI is just a
840 string, so it can be passed around through email or other out-of-band
844 def get_readonly_uri():
846 The dirnode ('1') URI returned by this method can be used in
847 set_uri() on a different directory ('2') to 'mount' a reference to
848 this directory ('1') under the other ('2'). This URI is just a
849 string, so it can be passed around through email or other out-of-band
854 """I return a Deferred that fires with a dictionary mapping child
855 name (a unicode string) to (node, metadata_dict) tuples, in which
856 'node' is an IFilesystemNode (either IFileNode or IDirectoryNode),
857 and 'metadata_dict' is a dictionary of metadata."""
860 """I return a Deferred that fires with a boolean, True if there
861 exists a child of the given name, False if not. The child name must
862 be a unicode string."""
865 """I return a Deferred that fires with a specific named child node,
866 either an IFileNode or an IDirectoryNode. The child name must be a
867 unicode string. I raise NoSuchChildError if I do not have a child by
870 def get_metadata_for(name):
871 """I return a Deferred that fires with the metadata dictionary for a
872 specific named child node. This metadata is stored in the *edge*, not
873 in the child, so it is attached to the parent dirnode rather than the
874 child dir-or-file-node. The child name must be a unicode string. I
875 raise NoSuchChildError if I do not have a child by that name."""
877 def set_metadata_for(name, metadata):
878 """I replace any existing metadata for the named child with the new
879 metadata. The child name must be a unicode string. This metadata is
880 stored in the *edge*, not in the child, so it is attached to the
881 parent dirnode rather than the child dir-or-file-node. I return a
882 Deferred (that fires with this dirnode) when the operation is
883 complete. I raise NoSuchChildError if I do not have a child by that
886 def get_child_at_path(path):
887 """Transform a child path into an IFilesystemNode.
889 I perform a recursive series of 'get' operations to find the named
890 descendant node. I return a Deferred that fires with the node, or
891 errbacks with NoSuchChildError if the node could not be found.
893 The path can be either a single string (slash-separated) or a list of
894 path-name elements. All elements must be unicode strings.
897 def get_child_and_metadata_at_path(path):
898 """Transform a child path into an IFilesystemNode and metadata.
900 I am like get_child_at_path(), but my Deferred fires with a tuple of
901 (node, metadata). The metadata comes from the last edge. If the path
902 is empty, the metadata will be an empty dictionary.
905 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
906 """I add a child (by writecap+readcap) at the specific name. I return
907 a Deferred that fires when the operation finishes. If overwrite= is
908 True, I will replace any existing child of the same name, otherwise
909 an existing child will cause me to return ExistingChildError. The
910 child name must be a unicode string.
912 The child caps could be for a file, or for a directory. If the new
913 child is read/write, you will provide both writecap and readcap. If
914 the child is read-only, you will provide the readcap write (i.e. the
915 writecap= and readcap= arguments will both be the child's readcap).
916 The filecaps are typically obtained from an IFilesystemNode with
917 get_uri() and get_readonly_uri().
919 If metadata= is provided, I will use it as the metadata for the named
920 edge. This will replace any existing metadata. If metadata= is left
921 as the default value of None, I will set ['mtime'] to the current
922 time, and I will set ['ctime'] to the current time if there was not
923 already a child by this name present. This roughly matches the
924 ctime/mtime semantics of traditional filesystems.
926 If this directory node is read-only, the Deferred will errback with a
927 NotWriteableError."""
929 def set_children(entries, overwrite=True):
930 """Add multiple children (by writecap+readcap) to a directory node.
931 Takes a dictionary, with childname as keys and (writecap, readcap)
932 tuples (or (writecap, readcap, metadata) triples) as values. Returns
933 a Deferred that fires (with this dirnode) when the operation
934 finishes. This is equivalent to calling set_uri() multiple times, but
935 is much more efficient. All child names must be unicode strings.
938 def set_node(name, child, metadata=None, overwrite=True):
939 """I add a child at the specific name. I return a Deferred that fires
940 when the operation finishes. This Deferred will fire with the child
941 node that was just added. I will replace any existing child of the
942 same name. The child name must be a unicode string. The 'child'
943 instance must be an instance providing IFilesystemNode.
945 If metadata= is provided, I will use it as the metadata for the named
946 edge. This will replace any existing metadata. If metadata= is left
947 as the default value of None, I will set ['mtime'] to the current
948 time, and I will set ['ctime'] to the current time if there was not
949 already a child by this name present. This roughly matches the
950 ctime/mtime semantics of traditional filesystems.
952 If this directory node is read-only, the Deferred will errback with a
953 NotWriteableError."""
955 def set_nodes(entries, overwrite=True):
956 """Add multiple children to a directory node. Takes a dict mapping
957 unicode childname to (child_node, metdata) tuples. If metdata=None,
958 the original metadata is left unmodified. Returns a Deferred that
959 fires (with this dirnode) when the operation finishes. This is
960 equivalent to calling set_node() multiple times, but is much more
963 def add_file(name, uploadable, metadata=None, overwrite=True):
964 """I upload a file (using the given IUploadable), then attach the
965 resulting ImmutableFileNode to the directory at the given name. I set
966 metadata the same way as set_uri and set_node. The child name must be
969 I return a Deferred that fires (with the IFileNode of the uploaded
970 file) when the operation completes."""
973 """I remove the child at the specific name. I return a Deferred that
974 fires when the operation finishes. The child name must be a unicode
975 string. I raise NoSuchChildError if I do not have a child by that
978 def create_subdirectory(name, initial_children={}, overwrite=True):
979 """I create and attach a directory at the given name. The new
980 directory can be empty, or it can be populated with children
981 according to 'initial_children', which takes a dictionary in the same
982 format as set_nodes (i.e. mapping unicode child name to (childnode,
983 metadata) tuples). The child name must be a unicode string. I return
984 a Deferred that fires (with the new directory node) when the
985 operation finishes."""
987 def move_child_to(current_child_name, new_parent, new_child_name=None,
989 """I take one of my children and move them to a new parent. The child
990 is referenced by name. On the new parent, the child will live under
991 'new_child_name', which defaults to 'current_child_name'. TODO: what
992 should we do about metadata? I return a Deferred that fires when the
993 operation finishes. The child name must be a unicode string. I raise
994 NoSuchChildError if I do not have a child by that name."""
996 def build_manifest():
997 """I generate a table of everything reachable from this directory.
998 I also compute deep-stats as described below.
1000 I return a Monitor. The Monitor's results will be a dictionary with
1003 res['manifest']: a list of (path, cap) tuples for all nodes
1004 (directories and files) reachable from this one.
1005 'path' will be a tuple of unicode strings. The
1006 origin dirnode will be represented by an empty path
1008 res['verifycaps']: a list of (printable) verifycap strings, one for
1009 each reachable non-LIT node. This is a set:
1010 it will contain no duplicates.
1011 res['storage-index']: a list of (base32) storage index strings,
1012 one for each reachable non-LIT node. This is
1013 a set: it will contain no duplicates.
1014 res['stats']: a dictionary, the same that is generated by
1015 start_deep_stats() below.
1017 The Monitor will also have an .origin_si attribute with the (binary)
1018 storage index of the starting point.
1021 def start_deep_stats():
1022 """Return a Monitor, examining all nodes (directories and files)
1023 reachable from this one. The Monitor's results will be a dictionary
1024 with the following keys::
1026 count-immutable-files: count of how many CHK files are in the set
1027 count-mutable-files: same, for mutable files (does not include
1029 count-literal-files: same, for LIT files
1030 count-files: sum of the above three
1032 count-directories: count of directories
1034 size-immutable-files: total bytes for all CHK files in the set
1035 size-mutable-files (TODO): same, for current version of all mutable
1036 files, does not include directories
1037 size-literal-files: same, for LIT files
1038 size-directories: size of mutable files used by directories
1040 largest-directory: number of bytes in the largest directory
1041 largest-directory-children: number of children in the largest
1043 largest-immutable-file: number of bytes in the largest CHK file
1045 size-mutable-files is not yet implemented, because it would involve
1046 even more queries than deep_stats does.
1048 The Monitor will also have an .origin_si attribute with the (binary)
1049 storage index of the starting point.
1051 This operation will visit every directory node underneath this one,
1052 and can take a long time to run. On a typical workstation with good
1053 bandwidth, this can examine roughly 15 directories per second (and
1054 takes several minutes of 100% CPU for ~1700 directories).
1057 class ICodecEncoder(Interface):
1058 def set_params(data_size, required_shares, max_shares):
1059 """Set up the parameters of this encoder.
1061 This prepares the encoder to perform an operation that converts a
1062 single block of data into a number of shares, such that a future
1063 ICodecDecoder can use a subset of these shares to recover the
1064 original data. This operation is invoked by calling encode(). Once
1065 the encoding parameters are set up, the encode operation can be
1066 invoked multiple times.
1068 set_params() prepares the encoder to accept blocks of input data that
1069 are exactly 'data_size' bytes in length. The encoder will be prepared
1070 to produce 'max_shares' shares for each encode() operation (although
1071 see the 'desired_share_ids' to use less CPU). The encoding math will
1072 be chosen such that the decoder can get by with as few as
1073 'required_shares' of these shares and still reproduce the original
1074 data. For example, set_params(1000, 5, 5) offers no redundancy at
1075 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1077 Numerical Restrictions: 'data_size' is required to be an integral
1078 multiple of 'required_shares'. In general, the caller should choose
1079 required_shares and max_shares based upon their reliability
1080 requirements and the number of peers available (the total storage
1081 space used is roughly equal to max_shares*data_size/required_shares),
1082 then choose data_size to achieve the memory footprint desired (larger
1083 data_size means more efficient operation, smaller data_size means
1084 smaller memory footprint).
1086 In addition, 'max_shares' must be equal to or greater than
1087 'required_shares'. Of course, setting them to be equal causes
1088 encode() to degenerate into a particularly slow form of the 'split'
1091 See encode() for more details about how these parameters are used.
1093 set_params() must be called before any other ICodecEncoder methods
1098 """Return the 3-tuple of data_size, required_shares, max_shares"""
1100 def get_encoder_type():
1101 """Return a short string that describes the type of this encoder.
1103 There is required to be a global table of encoder classes. This method
1104 returns an index into this table; the value at this index is an
1105 encoder class, and this encoder is an instance of that class.
1108 def get_block_size():
1109 """Return the length of the shares that encode() will produce.
1112 def encode_proposal(data, desired_share_ids=None):
1113 """Encode some data.
1115 'data' must be a string (or other buffer object), and len(data) must
1116 be equal to the 'data_size' value passed earlier to set_params().
1118 This will return a Deferred that will fire with two lists. The first
1119 is a list of shares, each of which is a string (or other buffer
1120 object) such that len(share) is the same as what get_share_size()
1121 returned earlier. The second is a list of shareids, in which each is
1122 an integer. The lengths of the two lists will always be equal to each
1123 other. The user should take care to keep each share closely
1124 associated with its shareid, as one is useless without the other.
1126 The length of this output list will normally be the same as the value
1127 provided to the 'max_shares' parameter of set_params(). This may be
1128 different if 'desired_share_ids' is provided.
1130 'desired_share_ids', if provided, is required to be a sequence of
1131 ints, each of which is required to be >= 0 and < max_shares. If not
1132 provided, encode() will produce 'max_shares' shares, as if
1133 'desired_share_ids' were set to range(max_shares). You might use this
1134 if you initially thought you were going to use 10 peers, started
1135 encoding, and then two of the peers dropped out: you could use
1136 desired_share_ids= to skip the work (both memory and CPU) of
1137 producing shares for the peers which are no longer available.
1141 def encode(inshares, desired_share_ids=None):
1142 """Encode some data. This may be called multiple times. Each call is
1145 inshares is a sequence of length required_shares, containing buffers
1146 (i.e. strings), where each buffer contains the next contiguous
1147 non-overlapping segment of the input data. Each buffer is required to
1148 be the same length, and the sum of the lengths of the buffers is
1149 required to be exactly the data_size promised by set_params(). (This
1150 implies that the data has to be padded before being passed to
1151 encode(), unless of course it already happens to be an even multiple
1152 of required_shares in length.)
1154 ALSO: the requirement to break up your data into 'required_shares'
1155 chunks before calling encode() feels a bit surprising, at least from
1156 the point of view of a user who doesn't know how FEC works. It feels
1157 like an implementation detail that has leaked outside the
1158 abstraction barrier. Can you imagine a use case in which the data to
1159 be encoded might already be available in pre-segmented chunks, such
1160 that it is faster or less work to make encode() take a list rather
1161 than splitting a single string?
1163 ALSO ALSO: I think 'inshares' is a misleading term, since encode()
1164 is supposed to *produce* shares, so what it *accepts* should be
1165 something other than shares. Other places in this interface use the
1166 word 'data' for that-which-is-not-shares.. maybe we should use that
1169 'desired_share_ids', if provided, is required to be a sequence of
1170 ints, each of which is required to be >= 0 and < max_shares. If not
1171 provided, encode() will produce 'max_shares' shares, as if
1172 'desired_share_ids' were set to range(max_shares). You might use this
1173 if you initially thought you were going to use 10 peers, started
1174 encoding, and then two of the peers dropped out: you could use
1175 desired_share_ids= to skip the work (both memory and CPU) of
1176 producing shares for the peers which are no longer available.
1178 For each call, encode() will return a Deferred that fires with two
1179 lists, one containing shares and the other containing the shareids.
1180 The get_share_size() method can be used to determine the length of
1181 the share strings returned by encode(). Each shareid is a small
1182 integer, exactly as passed into 'desired_share_ids' (or
1183 range(max_shares), if desired_share_ids was not provided).
1185 The shares and their corresponding shareids are required to be kept
1186 together during storage and retrieval. Specifically, the share data is
1187 useless by itself: the decoder needs to be told which share is which
1188 by providing it with both the shareid and the actual share data.
1190 This function will allocate an amount of memory roughly equal to::
1192 (max_shares - required_shares) * get_share_size()
1194 When combined with the memory that the caller must allocate to
1195 provide the input data, this leads to a memory footprint roughly
1196 equal to the size of the resulting encoded shares (i.e. the expansion
1197 factor times the size of the input segment).
1202 # returning a list of (shareidN,shareN) tuples instead of a pair of
1203 # lists (shareids..,shares..). Brian thought the tuples would
1204 # encourage users to keep the share and shareid together throughout
1205 # later processing, Zooko pointed out that the code to iterate
1206 # through two lists is not really more complicated than using a list
1207 # of tuples and there's also a performance improvement
1209 # having 'data_size' not required to be an integral multiple of
1210 # 'required_shares'. Doing this would require encode() to perform
1211 # padding internally, and we'd prefer to have any padding be done
1212 # explicitly by the caller. Yes, it is an abstraction leak, but
1213 # hopefully not an onerous one.
1216 class ICodecDecoder(Interface):
1217 def set_params(data_size, required_shares, max_shares):
1218 """Set the params. They have to be exactly the same ones that were
1219 used for encoding."""
1221 def get_needed_shares():
1222 """Return the number of shares needed to reconstruct the data.
1223 set_params() is required to be called before this."""
1225 def decode(some_shares, their_shareids):
1226 """Decode a partial list of shares into data.
1228 'some_shares' is required to be a sequence of buffers of sharedata, a
1229 subset of the shares returned by ICodecEncode.encode(). Each share is
1230 required to be of the same length. The i'th element of their_shareids
1231 is required to be the shareid of the i'th buffer in some_shares.
1233 This returns a Deferred which fires with a sequence of buffers. This
1234 sequence will contain all of the segments of the original data, in
1235 order. The sum of the lengths of all of the buffers will be the
1236 'data_size' value passed into the original ICodecEncode.set_params()
1237 call. To get back the single original input block of data, use
1238 ''.join(output_buffers), or you may wish to simply write them in
1239 order to an output file.
1241 Note that some of the elements in the result sequence may be
1242 references to the elements of the some_shares input sequence. In
1243 particular, this means that if those share objects are mutable (e.g.
1244 arrays) and if they are changed, then both the input (the
1245 'some_shares' parameter) and the output (the value given when the
1246 deferred is triggered) will change.
1248 The length of 'some_shares' is required to be exactly the value of
1249 'required_shares' passed into the original ICodecEncode.set_params()
1253 class IEncoder(Interface):
1254 """I take an object that provides IEncryptedUploadable, which provides
1255 encrypted data, and a list of shareholders. I then encode, hash, and
1256 deliver shares to those shareholders. I will compute all the necessary
1257 Merkle hash trees that are necessary to validate the crypttext that
1258 eventually comes back from the shareholders. I provide the URI Extension
1259 Block Hash, and the encoding parameters, both of which must be included
1262 I do not choose shareholders, that is left to the IUploader. I must be
1263 given a dict of RemoteReferences to storage buckets that are ready and
1264 willing to receive data.
1268 """Specify the number of bytes that will be encoded. This must be
1269 peformed before get_serialized_params() can be called.
1271 def set_params(params):
1272 """Override the default encoding parameters. 'params' is a tuple of
1273 (k,d,n), where 'k' is the number of required shares, 'd' is the
1274 shares_of_happiness, and 'n' is the total number of shares that will
1277 Encoding parameters can be set in three ways. 1: The Encoder class
1278 provides defaults (3/7/10). 2: the Encoder can be constructed with
1279 an 'options' dictionary, in which the
1280 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1281 set_params((k,d,n)) can be called.
1283 If you intend to use set_params(), you must call it before
1284 get_share_size or get_param are called.
1287 def set_encrypted_uploadable(u):
1288 """Provide a source of encrypted upload data. 'u' must implement
1289 IEncryptedUploadable.
1291 When this is called, the IEncryptedUploadable will be queried for its
1292 length and the storage_index that should be used.
1294 This returns a Deferred that fires with this Encoder instance.
1296 This must be performed before start() can be called.
1299 def get_param(name):
1300 """Return an encoding parameter, by name.
1302 'storage_index': return a string with the (16-byte truncated SHA-256
1303 hash) storage index to which these shares should be
1306 'share_counts': return a tuple describing how many shares are used:
1307 (needed_shares, shares_of_happiness, total_shares)
1309 'num_segments': return an int with the number of segments that
1312 'segment_size': return an int with the size of each segment.
1314 'block_size': return the size of the individual blocks that will
1315 be delivered to a shareholder's put_block() method. By
1316 knowing this, the shareholder will be able to keep all
1317 blocks in a single file and still provide random access
1318 when reading them. # TODO: can we avoid exposing this?
1320 'share_size': an int with the size of the data that will be stored
1321 on each shareholder. This is aggregate amount of data
1322 that will be sent to the shareholder, summed over all
1323 the put_block() calls I will ever make. It is useful to
1324 determine this size before asking potential
1325 shareholders whether they will grant a lease or not,
1326 since their answers will depend upon how much space we
1327 need. TODO: this might also include some amount of
1328 overhead, like the size of all the hashes. We need to
1329 decide whether this is useful or not.
1331 'serialized_params': a string with a concise description of the
1332 codec name and its parameters. This may be passed
1333 into the IUploadable to let it make sure that
1334 the same file encoded with different parameters
1335 will result in different storage indexes.
1337 Once this is called, set_size() and set_params() may not be called.
1340 def set_shareholders(shareholders):
1341 """Tell the encoder where to put the encoded shares. 'shareholders'
1342 must be a dictionary that maps share number (an integer ranging from
1343 0 to n-1) to an instance that provides IStorageBucketWriter. This
1344 must be performed before start() can be called."""
1347 """Begin the encode/upload process. This involves reading encrypted
1348 data from the IEncryptedUploadable, encoding it, uploading the shares
1349 to the shareholders, then sending the hash trees.
1351 set_encrypted_uploadable() and set_shareholders() must be called
1352 before this can be invoked.
1354 This returns a Deferred that fires with a verify cap when the upload
1355 process is complete. The verifycap, plus the encryption key, is
1356 sufficient to construct the read cap.
1359 class IDecoder(Interface):
1360 """I take a list of shareholders and some setup information, then
1361 download, validate, decode, and decrypt data from them, writing the
1362 results to an output file.
1364 I do not locate the shareholders, that is left to the IDownloader. I must
1365 be given a dict of RemoteReferences to storage buckets that are ready to
1370 """I take a file-like object (providing write and close) to which all
1371 the plaintext data will be written.
1373 TODO: producer/consumer . Maybe write() should return a Deferred that
1374 indicates when it will accept more data? But probably having the
1375 IDecoder be a producer is easier to glue to IConsumer pieces.
1378 def set_shareholders(shareholders):
1379 """I take a dictionary that maps share identifiers (small integers)
1380 to RemoteReferences that provide RIBucketReader. This must be called
1384 """I start the download. This process involves retrieving data and
1385 hash chains from the shareholders, using the hashes to validate the
1386 data, decoding the shares into segments, decrypting the segments,
1387 then writing the resulting plaintext to the output file.
1389 I return a Deferred that will fire (with self) when the download is
1393 class IDownloadTarget(Interface):
1394 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1395 # will register itself as a producer. This allows the target to invoke
1396 # downloader.pauseProducing, resumeProducing, and stopProducing.
1398 """Called before any calls to write() or close(). If an error
1399 occurs before any data is available, fail() may be called without
1400 a previous call to open().
1402 'size' is the length of the file being downloaded, in bytes."""
1405 """Output some data to the target."""
1407 """Inform the target that there is no more data to be written."""
1409 """fail() is called to indicate that the download has failed. 'why'
1410 is a Failure object indicating what went wrong. No further methods
1411 will be invoked on the IDownloadTarget after fail()."""
1412 def register_canceller(cb):
1413 """The CiphertextDownloader uses this to register a no-argument function
1414 that the target can call to cancel the download. Once this canceller
1415 is invoked, no further calls to write() or close() will be made."""
1417 """When the CiphertextDownloader is done, this finish() function will be
1418 called. Whatever it returns will be returned to the invoker of
1419 Downloader.download.
1421 # The following methods are just because that target might be a
1422 # repairer.DownUpConnector, and just because the current CHKUpload object
1423 # expects to find the storage index and encoding parameters in its
1425 def set_storageindex(storageindex):
1426 """ Set the storage index. """
1427 def set_encodingparams(encodingparams):
1428 """ Set the encoding parameters. """
1430 class IDownloader(Interface):
1431 def download(uri, target):
1432 """Perform a CHK download, sending the data to the given target.
1433 'target' must provide IDownloadTarget.
1435 Returns a Deferred that fires (with the results of target.finish)
1436 when the download is finished, or errbacks if something went wrong."""
1438 class IEncryptedUploadable(Interface):
1439 def set_upload_status(upload_status):
1440 """Provide an IUploadStatus object that should be filled with status
1441 information. The IEncryptedUploadable is responsible for setting
1442 key-determination progress ('chk'), size, storage_index, and
1443 ciphertext-fetch progress. It may delegate some of this
1444 responsibility to others, in particular to the IUploadable."""
1447 """This behaves just like IUploadable.get_size()."""
1449 def get_all_encoding_parameters():
1450 """Return a Deferred that fires with a tuple of
1451 (k,happy,n,segment_size). The segment_size will be used as-is, and
1452 must match the following constraints: it must be a multiple of k, and
1453 it shouldn't be unreasonably larger than the file size (if
1454 segment_size is larger than filesize, the difference must be stored
1457 This usually passes through to the IUploadable method of the same
1460 The encoder strictly obeys the values returned by this method. To
1461 make an upload use non-default encoding parameters, you must arrange
1462 to control the values that this method returns.
1465 def get_storage_index():
1466 """Return a Deferred that fires with a 16-byte storage index.
1469 def read_encrypted(length, hash_only):
1470 """This behaves just like IUploadable.read(), but returns crypttext
1471 instead of plaintext. If hash_only is True, then this discards the
1472 data (and returns an empty list); this improves efficiency when
1473 resuming an interrupted upload (where we need to compute the
1474 plaintext hashes, but don't need the redundant encrypted data)."""
1476 def get_plaintext_hashtree_leaves(first, last, num_segments):
1477 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1478 plaintext segments, i.e. get the tagged hashes of the given segments.
1479 The segment size is expected to be generated by the
1480 IEncryptedUploadable before any plaintext is read or ciphertext
1481 produced, so that the segment hashes can be generated with only a
1484 This returns a Deferred which fires with a sequence of hashes, using:
1486 tuple(segment_hashes[first:last])
1488 'num_segments' is used to assert that the number of segments that the
1489 IEncryptedUploadable handled matches the number of segments that the
1490 encoder was expecting.
1492 This method must not be called until the final byte has been read
1493 from read_encrypted(). Once this method is called, read_encrypted()
1494 can never be called again.
1497 def get_plaintext_hash():
1498 """OBSOLETE; Get the hash of the whole plaintext.
1500 This returns a Deferred which fires with a tagged SHA-256 hash of the
1501 whole plaintext, obtained from hashutil.plaintext_hash(data).
1505 """Just like IUploadable.close()."""
1507 class IUploadable(Interface):
1508 def set_upload_status(upload_status):
1509 """Provide an IUploadStatus object that should be filled with status
1510 information. The IUploadable is responsible for setting
1511 key-determination progress ('chk')."""
1513 def set_default_encoding_parameters(params):
1514 """Set the default encoding parameters, which must be a dict mapping
1515 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1516 'max_segment_size'. These might have an influence on the final
1517 encoding parameters returned by get_all_encoding_parameters(), if the
1518 Uploadable doesn't have more specific preferences.
1520 This call is optional: if it is not used, the Uploadable will use
1521 some built-in defaults. If used, this method must be called before
1522 any other IUploadable methods to have any effect.
1526 """Return a Deferred that will fire with the length of the data to be
1527 uploaded, in bytes. This will be called before the data is actually
1528 used, to compute encoding parameters.
1531 def get_all_encoding_parameters():
1532 """Return a Deferred that fires with a tuple of
1533 (k,happy,n,segment_size). The segment_size will be used as-is, and
1534 must match the following constraints: it must be a multiple of k, and
1535 it shouldn't be unreasonably larger than the file size (if
1536 segment_size is larger than filesize, the difference must be stored
1539 The relative values of k and n allow some IUploadables to request
1540 better redundancy than others (in exchange for consuming more space
1543 Larger values of segment_size reduce hash overhead, while smaller
1544 values reduce memory footprint and cause data to be delivered in
1545 smaller pieces (which may provide a smoother and more predictable
1546 download experience).
1548 The encoder strictly obeys the values returned by this method. To
1549 make an upload use non-default encoding parameters, you must arrange
1550 to control the values that this method returns. One way to influence
1551 them may be to call set_encoding_parameters() before calling
1552 get_all_encoding_parameters().
1555 def get_encryption_key():
1556 """Return a Deferred that fires with a 16-byte AES key. This key will
1557 be used to encrypt the data. The key will also be hashed to derive
1560 Uploadables which want to achieve convergence should hash their file
1561 contents and the serialized_encoding_parameters to form the key
1562 (which of course requires a full pass over the data). Uploadables can
1563 use the upload.ConvergentUploadMixin class to achieve this
1566 Uploadables which do not care about convergence (or do not wish to
1567 make multiple passes over the data) can simply return a
1568 strongly-random 16 byte string.
1570 get_encryption_key() may be called multiple times: the IUploadable is
1571 required to return the same value each time.
1575 """Return a Deferred that fires with a list of strings (perhaps with
1576 only a single element) which, when concatenated together, contain the
1577 next 'length' bytes of data. If EOF is near, this may provide fewer
1578 than 'length' bytes. The total number of bytes provided by read()
1579 before it signals EOF must equal the size provided by get_size().
1581 If the data must be acquired through multiple internal read
1582 operations, returning a list instead of a single string may help to
1583 reduce string copies.
1585 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1586 so a 10kB file means length=3kB, 100kB file means length=30kB,
1587 and >=1MB file means length=300kB.
1589 This method provides for a single full pass through the data. Later
1590 use cases may desire multiple passes or access to only parts of the
1591 data (such as a mutable file making small edits-in-place). This API
1592 will be expanded once those use cases are better understood.
1596 """The upload is finished, and whatever filehandle was in use may be
1599 class IUploadResults(Interface):
1600 """I am returned by upload() methods. I contain a number of public
1601 attributes which can be read to determine the results of the upload. Some
1602 of these are functional, some are timing information. All of these may be
1605 .file_size : the size of the file, in bytes
1606 .uri : the CHK read-cap for the file
1607 .ciphertext_fetched : how many bytes were fetched by the helper
1608 .sharemap: dict mapping share identifier to set of serverids
1609 (binary strings). This indicates which servers were given
1610 which shares. For immutable files, the shareid is an
1611 integer (the share number, from 0 to N-1). For mutable
1612 files, it is a string of the form 'seq%d-%s-sh%d',
1613 containing the sequence number, the roothash, and the
1615 .servermap : dict mapping server peerid to a set of share numbers
1616 .timings : dict of timing information, mapping name to seconds (float)
1617 total : total upload time, start to finish
1618 storage_index : time to compute the storage index
1619 peer_selection : time to decide which peers will be used
1620 contacting_helper : initial helper query to upload/no-upload decision
1621 existence_check : helper pre-upload existence check
1622 helper_total : initial helper query to helper finished pushing
1623 cumulative_fetch : helper waiting for ciphertext requests
1624 total_fetch : helper start to last ciphertext response
1625 cumulative_encoding : just time spent in zfec
1626 cumulative_sending : just time spent waiting for storage servers
1627 hashes_and_close : last segment push to shareholder close
1628 total_encode_and_push : first encode to shareholder close
1632 class IDownloadResults(Interface):
1633 """I am created internally by download() methods. I contain a number of
1634 public attributes which contain details about the download process.::
1636 .file_size : the size of the file, in bytes
1637 .servers_used : set of server peerids that were used during download
1638 .server_problems : dict mapping server peerid to a problem string. Only
1639 servers that had problems (bad hashes, disconnects)
1641 .servermap : dict mapping server peerid to a set of share numbers. Only
1642 servers that had any shares are listed here.
1643 .timings : dict of timing information, mapping name to seconds (float)
1644 peer_selection : time to ask servers about shares
1645 servers_peer_selection : dict of peerid to DYHB-query time
1646 uri_extension : time to fetch a copy of the URI extension block
1647 hashtrees : time to fetch the hash trees
1648 segments : time to fetch, decode, and deliver segments
1649 cumulative_fetch : time spent waiting for storage servers
1650 cumulative_decode : just time spent in zfec
1651 cumulative_decrypt : just time spent in decryption
1652 total : total download time, start to finish
1653 fetch_per_server : dict of peerid to list of per-segment fetch times
1657 class IUploader(Interface):
1658 def upload(uploadable):
1659 """Upload the file. 'uploadable' must impement IUploadable. This
1660 returns a Deferred which fires with an IUploadResults instance, from
1661 which the URI of the file can be obtained as results.uri ."""
1663 def upload_ssk(write_capability, new_version, uploadable):
1664 """TODO: how should this work?"""
1666 class ICheckable(Interface):
1667 def check(monitor, verify=False, add_lease=False):
1668 """Check upon my health, optionally repairing any problems.
1670 This returns a Deferred that fires with an instance that provides
1671 ICheckResults, or None if the object is non-distributed (i.e. LIT
1674 The monitor will be checked periodically to see if the operation has
1675 been cancelled. If so, no new queries will be sent, and the Deferred
1676 will fire (with a OperationCancelledError) immediately.
1678 Filenodes and dirnodes (which provide IFilesystemNode) are also
1679 checkable. Instances that represent verifier-caps will be checkable
1680 but not downloadable. Some objects (like LIT files) do not actually
1681 live in the grid, and their checkers return None (non-distributed
1682 files are always healthy).
1684 If verify=False, a relatively lightweight check will be performed: I
1685 will ask all servers if they have a share for me, and I will believe
1686 whatever they say. If there are at least N distinct shares on the
1687 grid, my results will indicate r.is_healthy()==True. This requires a
1688 roundtrip to each server, but does not transfer very much data, so
1689 the network bandwidth is fairly low.
1691 If verify=True, a more resource-intensive check will be performed:
1692 every share will be downloaded, and the hashes will be validated on
1693 every bit. I will ignore any shares that failed their hash checks. If
1694 there are at least N distinct valid shares on the grid, my results
1695 will indicate r.is_healthy()==True. This requires N/k times as much
1696 download bandwidth (and server disk IO) as a regular download. If a
1697 storage server is holding a corrupt share, or is experiencing memory
1698 failures during retrieval, or is malicious or buggy, then
1699 verification will detect the problem, but checking will not.
1701 If add_lease=True, I will ensure that an up-to-date lease is present
1702 on each share. The lease secrets will be derived from by node secret
1703 (in BASEDIR/private/secret), so either I will add a new lease to the
1704 share, or I will merely renew the lease that I already had. In a
1705 future version of the storage-server protocol (once Accounting has
1706 been implemented), there may be additional options here to define the
1707 kind of lease that is obtained (which account number to claim, etc).
1709 TODO: any problems seen during checking will be reported to the
1710 health-manager.furl, a centralized object which is responsible for
1711 figuring out why files are unhealthy so corrective action can be
1715 def check_and_repair(monitor, verify=False, add_lease=False):
1716 """Like check(), but if the file/directory is not healthy, attempt to
1719 Any non-healthy result will cause an immediate repair operation, to
1720 generate and upload new shares. After repair, the file will be as
1721 healthy as we can make it. Details about what sort of repair is done
1722 will be put in the check-and-repair results. The Deferred will not
1723 fire until the repair is complete.
1725 This returns a Deferred which fires with an instance of
1726 ICheckAndRepairResults."""
1728 class IDeepCheckable(Interface):
1729 def start_deep_check(verify=False, add_lease=False):
1730 """Check upon the health of me and everything I can reach.
1732 This is a recursive form of check(), useable only on dirnodes.
1734 I return a Monitor, with results that are an IDeepCheckResults
1737 TODO: If any of the directories I traverse are unrecoverable, the
1738 Monitor will report failure. If any of the files I check upon are
1739 unrecoverable, those problems will be reported in the
1740 IDeepCheckResults as usual, and the Monitor will not report a
1744 def start_deep_check_and_repair(verify=False, add_lease=False):
1745 """Check upon the health of me and everything I can reach. Repair
1746 anything that isn't healthy.
1748 This is a recursive form of check_and_repair(), useable only on
1751 I return a Monitor, with results that are an
1752 IDeepCheckAndRepairResults object.
1754 TODO: If any of the directories I traverse are unrecoverable, the
1755 Monitor will report failure. If any of the files I check upon are
1756 unrecoverable, those problems will be reported in the
1757 IDeepCheckResults as usual, and the Monitor will not report a
1761 class ICheckResults(Interface):
1762 """I contain the detailed results of a check/verify operation.
1765 def get_storage_index():
1766 """Return a string with the (binary) storage index."""
1767 def get_storage_index_string():
1768 """Return a string with the (printable) abbreviated storage index."""
1770 """Return the (string) URI of the object that was checked."""
1773 """Return a boolean, True if the file/dir is fully healthy, False if
1774 it is damaged in any way. Non-distributed LIT files always return
1777 def is_recoverable():
1778 """Return a boolean, True if the file/dir can be recovered, False if
1779 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
1780 files always return True."""
1782 def needs_rebalancing():
1783 """Return a boolean, True if the file/dir's reliability could be
1784 improved by moving shares to new servers. Non-distributed LIT files
1785 always return False."""
1789 """Return a dictionary that describes the state of the file/dir. LIT
1790 files always return an empty dictionary. Normal files and directories
1791 return a dictionary with the following keys (note that these use
1792 binary strings rather than base32-encoded ones) (also note that for
1793 mutable files, these counts are for the 'best' version):
1795 count-shares-good: the number of distinct good shares that were found
1796 count-shares-needed: 'k', the number of shares required for recovery
1797 count-shares-expected: 'N', the number of total shares generated
1798 count-good-share-hosts: the number of distinct storage servers with
1799 good shares. If this number is less than
1800 count-shares-good, then some shares are
1801 doubled up, increasing the correlation of
1802 failures. This indicates that one or more
1803 shares should be moved to an otherwise unused
1804 server, if one is available.
1805 count-corrupt-shares: the number of shares with integrity failures
1806 list-corrupt-shares: a list of 'share locators', one for each share
1807 that was found to be corrupt. Each share
1808 locator is a list of (serverid, storage_index,
1810 count-incompatible-shares: the number of shares which are of a share
1811 format unknown to this checker
1812 list-incompatible-shares: a list of 'share locators', one for each
1813 share that was found to be of an unknown
1814 format. Each share locator is a list of
1815 (serverid, storage_index, sharenum).
1816 servers-responding: list of (binary) storage server identifiers,
1817 one for each server which responded to the share
1818 query (even if they said they didn't have
1819 shares, and even if they said they did have
1820 shares but then didn't send them when asked, or
1821 dropped the connection, or returned a Failure,
1822 and even if they said they did have shares and
1823 sent incorrect ones when asked)
1824 sharemap: dict mapping share identifier to list of serverids
1825 (binary strings). This indicates which servers are holding
1826 which shares. For immutable files, the shareid is an
1827 integer (the share number, from 0 to N-1). For mutable
1828 files, it is a string of the form 'seq%d-%s-sh%d',
1829 containing the sequence number, the roothash, and the
1832 The following keys are most relevant for mutable files, but immutable
1833 files will provide sensible values too::
1835 count-wrong-shares: the number of shares for versions other than the
1836 'best' one (which is defined as being the
1837 recoverable version with the highest sequence
1838 number, then the highest roothash). These are
1839 either leftover shares from an older version
1840 (perhaps on a server that was offline when an
1841 update occurred), shares from an unrecoverable
1842 newer version, or shares from an alternate
1843 current version that results from an
1844 uncoordinated write collision. For a healthy
1845 file, this will equal 0.
1847 count-recoverable-versions: the number of recoverable versions of
1848 the file. For a healthy file, this will
1851 count-unrecoverable-versions: the number of unrecoverable versions
1852 of the file. For a healthy file, this
1858 """Return a string with a brief (one-line) summary of the results."""
1861 """Return a list of strings with more detailed results."""
1863 class ICheckAndRepairResults(Interface):
1864 """I contain the detailed results of a check/verify/repair operation.
1866 The IFilesystemNode.check()/verify()/repair() methods all return
1867 instances that provide ICheckAndRepairResults.
1870 def get_storage_index():
1871 """Return a string with the (binary) storage index."""
1872 def get_storage_index_string():
1873 """Return a string with the (printable) abbreviated storage index."""
1874 def get_repair_attempted():
1875 """Return a boolean, True if a repair was attempted. We might not
1876 attempt to repair the file because it was healthy, or healthy enough
1877 (i.e. some shares were missing but not enough to exceed some
1878 threshold), or because we don't know how to repair this object."""
1879 def get_repair_successful():
1880 """Return a boolean, True if repair was attempted and the file/dir
1881 was fully healthy afterwards. False if no repair was attempted or if
1882 a repair attempt failed."""
1883 def get_pre_repair_results():
1884 """Return an ICheckResults instance that describes the state of the
1885 file/dir before any repair was attempted."""
1886 def get_post_repair_results():
1887 """Return an ICheckResults instance that describes the state of the
1888 file/dir after any repair was attempted. If no repair was attempted,
1889 the pre-repair and post-repair results will be identical."""
1892 class IDeepCheckResults(Interface):
1893 """I contain the results of a deep-check operation.
1895 This is returned by a call to ICheckable.deep_check().
1898 def get_root_storage_index_string():
1899 """Return the storage index (abbreviated human-readable string) of
1900 the first object checked."""
1902 """Return a dictionary with the following keys::
1904 count-objects-checked: count of how many objects were checked
1905 count-objects-healthy: how many of those objects were completely
1907 count-objects-unhealthy: how many were damaged in some way
1908 count-objects-unrecoverable: how many were unrecoverable
1909 count-corrupt-shares: how many shares were found to have
1910 corruption, summed over all objects
1914 def get_corrupt_shares():
1915 """Return a set of (serverid, storage_index, sharenum) for all shares
1916 that were found to be corrupt. Both serverid and storage_index are
1919 def get_all_results():
1920 """Return a dictionary mapping pathname (a tuple of strings, ready to
1921 be slash-joined) to an ICheckResults instance, one for each object
1922 that was checked."""
1924 def get_results_for_storage_index(storage_index):
1925 """Retrive the ICheckResults instance for the given (binary)
1926 storage index. Raises KeyError if there are no results for that
1930 """Return a dictionary with the same keys as
1931 IDirectoryNode.deep_stats()."""
1933 class IDeepCheckAndRepairResults(Interface):
1934 """I contain the results of a deep-check-and-repair operation.
1936 This is returned by a call to ICheckable.deep_check_and_repair().
1939 def get_root_storage_index_string():
1940 """Return the storage index (abbreviated human-readable string) of
1941 the first object checked."""
1943 """Return a dictionary with the following keys::
1945 count-objects-checked: count of how many objects were checked
1946 count-objects-healthy-pre-repair: how many of those objects were
1947 completely healthy (before any
1949 count-objects-unhealthy-pre-repair: how many were damaged in
1951 count-objects-unrecoverable-pre-repair: how many were unrecoverable
1952 count-objects-healthy-post-repair: how many of those objects were
1953 completely healthy (after any
1955 count-objects-unhealthy-post-repair: how many were damaged in
1957 count-objects-unrecoverable-post-repair: how many were
1959 count-repairs-attempted: repairs were attempted on this many
1960 objects. The count-repairs- keys will
1961 always be provided, however unless
1962 repair=true is present, they will all
1964 count-repairs-successful: how many repairs resulted in healthy
1966 count-repairs-unsuccessful: how many repairs resulted did not
1967 results in completely healthy objects
1968 count-corrupt-shares-pre-repair: how many shares were found to
1969 have corruption, summed over all
1970 objects examined (before any
1972 count-corrupt-shares-post-repair: how many shares were found to
1973 have corruption, summed over all
1974 objects examined (after any
1979 """Return a dictionary with the same keys as
1980 IDirectoryNode.deep_stats()."""
1982 def get_corrupt_shares():
1983 """Return a set of (serverid, storage_index, sharenum) for all shares
1984 that were found to be corrupt before any repair was attempted. Both
1985 serverid and storage_index are binary.
1987 def get_remaining_corrupt_shares():
1988 """Return a set of (serverid, storage_index, sharenum) for all shares
1989 that were found to be corrupt after any repair was completed. Both
1990 serverid and storage_index are binary. These are shares that need
1991 manual inspection and probably deletion.
1993 def get_all_results():
1994 """Return a dictionary mapping pathname (a tuple of strings, ready to
1995 be slash-joined) to an ICheckAndRepairResults instance, one for each
1996 object that was checked."""
1998 def get_results_for_storage_index(storage_index):
1999 """Retrive the ICheckAndRepairResults instance for the given (binary)
2000 storage index. Raises KeyError if there are no results for that
2004 class IRepairable(Interface):
2005 def repair(check_results):
2006 """Attempt to repair the given object. Returns a Deferred that fires
2007 with a IRepairResults object.
2009 I must be called with an object that implements ICheckResults, as
2010 proof that you have actually discovered a problem with this file. I
2011 will use the data in the checker results to guide the repair process,
2012 such as which servers provided bad data and should therefore be
2013 avoided. The ICheckResults object is inside the
2014 ICheckAndRepairResults object, which is returned by the
2015 ICheckable.check() method::
2017 d = filenode.check(repair=False)
2018 def _got_results(check_and_repair_results):
2019 check_results = check_and_repair_results.get_pre_repair_results()
2020 return filenode.repair(check_results)
2021 d.addCallback(_got_results)
2025 class IRepairResults(Interface):
2026 """I contain the results of a repair operation."""
2027 def get_successful(self):
2028 """Returns a boolean: True if the repair made the file healthy, False
2029 if not. Repair failure generally indicates a file that has been
2030 damaged beyond repair."""
2033 class IClient(Interface):
2034 def upload(uploadable):
2035 """Upload some data into a CHK, get back the UploadResults for it.
2036 @param uploadable: something that implements IUploadable
2037 @return: a Deferred that fires with the UploadResults instance.
2038 To get the URI for this file, use results.uri .
2041 def create_mutable_file(contents=""):
2042 """Create a new mutable file (with initial) contents, get back the
2045 @param contents: (bytestring, callable, or None): this provides the
2046 initial contents of the mutable file. If 'contents' is a bytestring,
2047 it will be used as-is. If 'contents' is a callable, it will be
2048 invoked with the new MutableFileNode instance and is expected to
2049 return a bytestring with the initial contents of the file (the
2050 callable can use node.get_writekey() to decide how to encrypt the
2051 initial contents, e.g. for a brand new dirnode with initial
2052 children). contents=None is equivalent to an empty string. Using
2053 content_maker= is more efficient than creating a mutable file and
2054 setting its contents in two separate operations.
2056 @return: a Deferred that fires with an IMutableFileNode instance.
2059 def create_dirnode(initial_children={}):
2060 """Create a new unattached dirnode, possibly with initial children.
2062 @param initial_children: dict with keys that are unicode child names,
2063 and values that are (childnode, metadata) tuples.
2065 @return: a Deferred that fires with the new IDirectoryNode instance.
2068 def create_node_from_uri(uri, rouri):
2069 """Create a new IFilesystemNode instance from the uri, synchronously.
2070 @param uri: a string or IURI-providing instance, or None. This could
2071 be for a LiteralFileNode, a CHK file node, a mutable file
2072 node, or a directory node
2073 @param rouri: a string or IURI-providing instance, or None. If the
2074 main uri is None, I will use the rouri instead. If I
2075 recognize the format of the main uri, I will ignore the
2076 rouri (because it can be derived from the writecap).
2078 @return: an instance that provides IFilesystemNode (or more usefully
2079 one of its subclasses). File-specifying URIs will result in
2080 IFileNode-providing instances, like ImmutableFileNode,
2081 LiteralFileNode, or MutableFileNode. Directory-specifying
2082 URIs will result in IDirectoryNode-providing instances, like
2086 class INodeMaker(Interface):
2087 """The NodeMaker is used to create IFilesystemNode instances. It can
2088 accept a filecap/dircap string and return the node right away. It can
2089 also create new nodes (i.e. upload a file, or create a mutable file)
2090 asynchronously. Once you have one of these nodes, you can use other
2091 methods to determine whether it is a file or directory, and to download
2092 or modify its contents.
2094 The NodeMaker encapsulates all the authorities that these
2095 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2096 Tahoe process will typically have a single NodeMaker, but unit tests may
2097 create simplified/mocked forms for testing purposes.
2099 def create_from_cap(writecap, readcap=None, **kwargs):
2100 """I create an IFilesystemNode from the given writecap/readcap. I can
2101 only provide nodes for existing file/directory objects: use my other
2102 methods to create new objects. I return synchronously."""
2104 def create_mutable_file(contents=None, keysize=None):
2105 """I create a new mutable file, and return a Deferred which will fire
2106 with the IMutableFileNode instance when it is ready. If contents= is
2107 provided (a bytestring), it will be used as the initial contents of
2108 the new file, otherwise the file will contain zero bytes. keysize= is
2109 for use by unit tests, to create mutable files that are smaller than
2112 def create_new_mutable_directory(initial_children={}):
2113 """I create a new mutable directory, and return a Deferred which will
2114 fire with the IDirectoryNode instance when it is ready. If
2115 initial_children= is provided (a dict mapping unicode child name to
2116 (childnode, metadata_dict) tuples), the directory will be populated
2117 with those children, otherwise it will be empty."""
2119 class IClientStatus(Interface):
2120 def list_all_uploads():
2121 """Return a list of uploader objects, one for each upload which
2122 currently has an object available (tracked with weakrefs). This is
2123 intended for debugging purposes."""
2124 def list_active_uploads():
2125 """Return a list of active IUploadStatus objects."""
2126 def list_recent_uploads():
2127 """Return a list of IUploadStatus objects for the most recently
2130 def list_all_downloads():
2131 """Return a list of downloader objects, one for each download which
2132 currently has an object available (tracked with weakrefs). This is
2133 intended for debugging purposes."""
2134 def list_active_downloads():
2135 """Return a list of active IDownloadStatus objects."""
2136 def list_recent_downloads():
2137 """Return a list of IDownloadStatus objects for the most recently
2138 started downloads."""
2140 class IUploadStatus(Interface):
2142 """Return a timestamp (float with seconds since epoch) indicating
2143 when the operation was started."""
2144 def get_storage_index():
2145 """Return a string with the (binary) storage index in use on this
2146 upload. Returns None if the storage index has not yet been
2149 """Return an integer with the number of bytes that will eventually
2150 be uploaded for this file. Returns None if the size is not yet known.
2153 """Return True if this upload is using a Helper, False if not."""
2155 """Return a string describing the current state of the upload
2158 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2159 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2160 made towards hashing the file to determine a CHK encryption key: if
2161 non-convergent encryption is in use, this will be trivial, otherwise
2162 the whole file must be hashed. 'ciphertext' describes how much of the
2163 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2164 uploads. 'encode_and_push' describes how much of the encode-and-push
2165 process has finished: for helper uploads this is dependent upon the
2166 helper providing progress reports. It might be reasonable to add all
2167 three numbers and report the sum to the user."""
2169 """Return True if the upload is currently active, False if not."""
2171 """Return an instance of UploadResults (which contains timing and
2172 sharemap information). Might return None if the upload is not yet
2175 """Each upload status gets a unique number: this method returns that
2176 number. This provides a handle to this particular upload, so a web
2177 page can generate a suitable hyperlink."""
2179 class IDownloadStatus(Interface):
2181 """Return a timestamp (float with seconds since epoch) indicating
2182 when the operation was started."""
2183 def get_storage_index():
2184 """Return a string with the (binary) storage index in use on this
2185 download. This may be None if there is no storage index (i.e. LIT
2188 """Return an integer with the number of bytes that will eventually be
2189 retrieved for this file. Returns None if the size is not yet known.
2192 """Return True if this download is using a Helper, False if not."""
2194 """Return a string describing the current state of the download
2197 """Returns a float (from 0.0 to 1.0) describing the amount of the
2198 download that has completed. This value will remain at 0.0 until the
2199 first byte of plaintext is pushed to the download target."""
2201 """Return True if the download is currently active, False if not."""
2203 """Each download status gets a unique number: this method returns
2204 that number. This provides a handle to this particular download, so a
2205 web page can generate a suitable hyperlink."""
2207 class IServermapUpdaterStatus(Interface):
2209 class IPublishStatus(Interface):
2211 class IRetrieveStatus(Interface):
2214 class NotCapableError(Exception):
2215 """You have tried to write to a read-only node."""
2217 class BadWriteEnablerError(Exception):
2220 class RIControlClient(RemoteInterface):
2222 def wait_for_client_connections(num_clients=int):
2223 """Do not return until we have connections to at least NUM_CLIENTS
2227 def upload_from_file_to_uri(filename=str,
2228 convergence=ChoiceOf(None,
2229 StringConstraint(2**20))):
2230 """Upload a file to the grid. This accepts a filename (which must be
2231 absolute) that points to a file on the node's local disk. The node will
2232 read the contents of this file, upload it to the grid, then return the
2233 URI at which it was uploaded. If convergence is None then a random
2234 encryption key will be used, else the plaintext will be hashed, then
2235 that hash will be mixed together with the "convergence" string to form
2240 def download_from_uri_to_file(uri=URI, filename=str):
2241 """Download a file from the grid, placing it on the node's local disk
2242 at the given filename (which must be absolute[?]). Returns the
2243 absolute filename where the file was written."""
2248 def get_memory_usage():
2249 """Return a dict describes the amount of memory currently in use. The
2250 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2251 measuring memory consupmtion in bytes."""
2252 return DictOf(str, int)
2254 def speed_test(count=int, size=int, mutable=Any()):
2255 """Write 'count' tempfiles to disk, all of the given size. Measure
2256 how long (in seconds) it takes to upload them all to the servers.
2257 Then measure how long it takes to download all of them. If 'mutable'
2258 is 'create', time creation of mutable files. If 'mutable' is
2259 'upload', then time access to the same mutable file instead of
2262 Returns a tuple of (upload_time, download_time).
2264 return (float, float)
2266 def measure_peer_response_time():
2267 """Send a short message to each connected peer, and measure the time
2268 it takes for them to respond to it. This is a rough measure of the
2269 application-level round trip time.
2271 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2274 return DictOf(Nodeid, float)
2276 UploadResults = Any() #DictOf(str, str)
2278 class RIEncryptedUploadable(RemoteInterface):
2279 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2284 def get_all_encoding_parameters():
2285 return (int, int, int, long)
2287 def read_encrypted(offset=Offset, length=ReadSize):
2294 class RICHKUploadHelper(RemoteInterface):
2295 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2299 Return a dictionary of version information.
2301 return DictOf(str, Any())
2303 def upload(reader=RIEncryptedUploadable):
2304 return UploadResults
2307 class RIHelper(RemoteInterface):
2308 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2312 Return a dictionary of version information.
2314 return DictOf(str, Any())
2316 def upload_chk(si=StorageIndex):
2317 """See if a file with a given storage index needs uploading. The
2318 helper will ask the appropriate storage servers to see if the file
2319 has already been uploaded. If so, the helper will return a set of
2320 'upload results' that includes whatever hashes are needed to build
2321 the read-cap, and perhaps a truncated sharemap.
2323 If the file has not yet been uploaded (or if it was only partially
2324 uploaded), the helper will return an empty upload-results dictionary
2325 and also an RICHKUploadHelper object that will take care of the
2326 upload process. The client should call upload() on this object and
2327 pass it a reference to an RIEncryptedUploadable object that will
2328 provide ciphertext. When the upload is finished, the upload() method
2329 will finish and return the upload results.
2331 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2334 class RIStatsProvider(RemoteInterface):
2335 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2337 Provides access to statistics and monitoring information.
2342 returns a dictionary containing 'counters' and 'stats', each a
2343 dictionary with string counter/stat name keys, and numeric values.
2344 counters are monotonically increasing measures of work done, and
2345 stats are instantaneous measures (potentially time averaged
2348 return DictOf(str, DictOf(str, ChoiceOf(float, int, long)))
2350 class RIStatsGatherer(RemoteInterface):
2351 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2353 Provides a monitoring service for centralised collection of stats
2356 def provide(provider=RIStatsProvider, nickname=str):
2358 @param provider: a stats collector instance which should be polled
2359 periodically by the gatherer to collect stats.
2360 @param nickname: a name useful to identify the provided client
2365 class IStatsProducer(Interface):
2368 returns a dictionary, with str keys representing the names of stats
2369 to be monitored, and numeric values.
2372 class RIKeyGenerator(RemoteInterface):
2373 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2375 Provides a service offering to make RSA key pairs.
2378 def get_rsa_key_pair(key_size=int):
2380 @param key_size: the size of the signature key.
2381 @return: tuple(verifying_key, signing_key)
2383 return TupleOf(str, str)
2386 class FileTooLargeError(Exception):
2389 class IValidatedThingProxy(Interface):
2391 """ Acquire a thing and validate it. Return a deferred which is
2392 eventually fired with self if the thing is valid or errbacked if it
2393 can't be acquired or validated."""
2395 class InsufficientVersionError(Exception):
2396 def __init__(self, needed, got):
2397 self.needed = needed
2400 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2403 class EmptyPathnameComponentError(Exception):
2404 """The webapi disallows empty pathname components."""