2 from zope.interface import Interface
3 from foolscap.api import StringConstraint, ListOf, TupleOf, SetOf, DictOf, \
4 ChoiceOf, IntegerConstraint, Any, RemoteInterface, Referenceable
12 Hash = StringConstraint(maxLength=HASH_SIZE,
13 minLength=HASH_SIZE)# binary format 32-byte SHA256 hash
14 Nodeid = StringConstraint(maxLength=20,
15 minLength=20) # binary format 20-byte SHA1 hash
16 FURL = StringConstraint(1000)
17 StorageIndex = StringConstraint(16)
18 URI = StringConstraint(300) # kind of arbitrary
20 MAX_BUCKETS = 256 # per peer -- zfec offers at most 256 shares per file
22 DEFAULT_MAX_SEGMENT_SIZE = 128*1024
24 ShareData = StringConstraint(None)
25 URIExtensionData = StringConstraint(1000)
26 Number = IntegerConstraint(8) # 2**(8*8) == 16EiB ~= 18e18 ~= 18 exabytes
28 ReadSize = int # the 'int' constraint is 2**31 == 2Gib -- large files are processed in not-so-large increments
29 WriteEnablerSecret = Hash # used to protect mutable bucket modifications
30 LeaseRenewSecret = Hash # used to protect bucket lease renewal requests
31 LeaseCancelSecret = Hash # used to protect bucket lease cancellation requests
33 class RIStubClient(RemoteInterface):
34 """Each client publishes a service announcement for a dummy object called
35 the StubClient. This object doesn't actually offer any services, but the
36 announcement helps the Introducer keep track of which clients are
37 subscribed (so the grid admin can keep track of things like the size of
38 the grid and the client versions in use. This is the (empty)
39 RemoteInterface for the StubClient."""
41 class RIBucketWriter(RemoteInterface):
42 """ Objects of this kind live on the server side. """
43 def write(offset=Offset, data=ShareData):
48 If the data that has been written is incomplete or inconsistent then
49 the server will throw the data away, else it will store it for future
55 """Abandon all the data that has been written.
59 class RIBucketReader(RemoteInterface):
60 def read(offset=Offset, length=ReadSize):
63 def advise_corrupt_share(reason=str):
64 """Clients who discover hash failures in shares that they have
65 downloaded from me will use this method to inform me about the
66 failures. I will record their concern so that my operator can
67 manually inspect the shares in question. I return None.
69 This is a wrapper around RIStorageServer.advise_corrupt_share(),
70 which is tied to a specific share, and therefore does not need the
71 extra share-identifying arguments. Please see that method for full
75 TestVector = ListOf(TupleOf(Offset, ReadSize, str, str))
76 # elements are (offset, length, operator, specimen)
77 # operator is one of "lt, le, eq, ne, ge, gt"
78 # nop always passes and is used to fetch data while writing.
79 # you should use length==len(specimen) for everything except nop
80 DataVector = ListOf(TupleOf(Offset, ShareData))
81 # (offset, data). This limits us to 30 writes of 1MiB each per call
82 TestAndWriteVectorsForShares = DictOf(int,
85 ChoiceOf(None, Offset), # new_length
87 ReadVector = ListOf(TupleOf(Offset, ReadSize))
88 ReadData = ListOf(ShareData)
89 # returns data[offset:offset+length] for each element of TestVector
91 class RIStorageServer(RemoteInterface):
92 __remote_name__ = "RIStorageServer.tahoe.allmydata.com"
96 Return a dictionary of version information.
98 return DictOf(str, Any())
100 def allocate_buckets(storage_index=StorageIndex,
101 renew_secret=LeaseRenewSecret,
102 cancel_secret=LeaseCancelSecret,
103 sharenums=SetOf(int, maxLength=MAX_BUCKETS),
104 allocated_size=Offset, canary=Referenceable):
106 @param storage_index: the index of the bucket to be created or
108 @param sharenums: these are the share numbers (probably between 0 and
109 99) that the sender is proposing to store on this
111 @param renew_secret: This is the secret used to protect bucket refresh
112 This secret is generated by the client and
113 stored for later comparison by the server. Each
114 server is given a different secret.
115 @param cancel_secret: Like renew_secret, but protects bucket decref.
116 @param canary: If the canary is lost before close(), the bucket is
118 @return: tuple of (alreadygot, allocated), where alreadygot is what we
119 already have and allocated is what we hereby agree to accept.
120 New leases are added for shares in both lists.
122 return TupleOf(SetOf(int, maxLength=MAX_BUCKETS),
123 DictOf(int, RIBucketWriter, maxKeys=MAX_BUCKETS))
125 def add_lease(storage_index=StorageIndex,
126 renew_secret=LeaseRenewSecret,
127 cancel_secret=LeaseCancelSecret):
129 Add a new lease on the given bucket. If the renew_secret matches an
130 existing lease, that lease will be renewed instead. If there is no
131 bucket for the given storage_index, return silently. (note that in
132 tahoe-1.3.0 and earlier, IndexError was raised if there was no
135 return Any() # returns None now, but future versions might change
137 def renew_lease(storage_index=StorageIndex, renew_secret=LeaseRenewSecret):
139 Renew the lease on a given bucket, resetting the timer to 31 days.
140 Some networks will use this, some will not. If there is no bucket for
141 the given storage_index, IndexError will be raised.
143 For mutable shares, if the given renew_secret does not match an
144 existing lease, IndexError will be raised with a note listing the
145 server-nodeids on the existing leases, so leases on migrated shares
146 can be renewed or cancelled. For immutable shares, IndexError
147 (without the note) will be raised.
151 def cancel_lease(storage_index=StorageIndex,
152 cancel_secret=LeaseCancelSecret):
154 Cancel the lease on a given bucket. If this was the last lease on the
155 bucket, the bucket will be deleted. If there is no bucket for the
156 given storage_index, IndexError will be raised.
158 For mutable shares, if the given cancel_secret does not match an
159 existing lease, IndexError will be raised with a note listing the
160 server-nodeids on the existing leases, so leases on migrated shares
161 can be renewed or cancelled. For immutable shares, IndexError
162 (without the note) will be raised.
166 def get_buckets(storage_index=StorageIndex):
167 return DictOf(int, RIBucketReader, maxKeys=MAX_BUCKETS)
171 def slot_readv(storage_index=StorageIndex,
172 shares=ListOf(int), readv=ReadVector):
173 """Read a vector from the numbered shares associated with the given
174 storage index. An empty shares list means to return data from all
175 known shares. Returns a dictionary with one key per share."""
176 return DictOf(int, ReadData) # shnum -> results
178 def slot_testv_and_readv_and_writev(storage_index=StorageIndex,
179 secrets=TupleOf(WriteEnablerSecret,
182 tw_vectors=TestAndWriteVectorsForShares,
185 """General-purpose test-and-set operation for mutable slots. Perform
186 a bunch of comparisons against the existing shares. If they all pass,
187 then apply a bunch of write vectors to those shares. Then use the
188 read vectors to extract data from all the shares and return the data.
190 This method is, um, large. The goal is to allow clients to update all
191 the shares associated with a mutable file in a single round trip.
193 @param storage_index: the index of the bucket to be created or
195 @param write_enabler: a secret that is stored along with the slot.
196 Writes are accepted from any caller who can
197 present the matching secret. A different secret
198 should be used for each slot*server pair.
199 @param renew_secret: This is the secret used to protect bucket refresh
200 This secret is generated by the client and
201 stored for later comparison by the server. Each
202 server is given a different secret.
203 @param cancel_secret: Like renew_secret, but protects bucket decref.
205 The 'secrets' argument is a tuple of (write_enabler, renew_secret,
206 cancel_secret). The first is required to perform any write. The
207 latter two are used when allocating new shares. To simply acquire a
208 new lease on existing shares, use an empty testv and an empty writev.
210 Each share can have a separate test vector (i.e. a list of
211 comparisons to perform). If all vectors for all shares pass, then all
212 writes for all shares are recorded. Each comparison is a 4-tuple of
213 (offset, length, operator, specimen), which effectively does a bool(
214 (read(offset, length)) OPERATOR specimen ) and only performs the
215 write if all these evaluate to True. Basic test-and-set uses 'eq'.
216 Write-if-newer uses a seqnum and (offset, length, 'lt', specimen).
217 Write-if-same-or-newer uses 'le'.
219 Reads from the end of the container are truncated, and missing shares
220 behave like empty ones, so to assert that a share doesn't exist (for
221 use when creating a new share), use (0, 1, 'eq', '').
223 The write vector will be applied to the given share, expanding it if
224 necessary. A write vector applied to a share number that did not
225 exist previously will cause that share to be created.
227 Each write vector is accompanied by a 'new_length' argument. If
228 new_length is not None, use it to set the size of the container. This
229 can be used to pre-allocate space for a series of upcoming writes, or
230 truncate existing data. If the container is growing, new_length will
231 be applied before datav. If the container is shrinking, it will be
232 applied afterwards. If new_length==0, the share will be deleted.
234 The read vector is used to extract data from all known shares,
235 *before* any writes have been applied. The same vector is used for
236 all shares. This captures the state that was tested by the test
239 This method returns two values: a boolean and a dict. The boolean is
240 True if the write vectors were applied, False if not. The dict is
241 keyed by share number, and each value contains a list of strings, one
242 for each element of the read vector.
244 If the write_enabler is wrong, this will raise BadWriteEnablerError.
245 To enable share migration (using update_write_enabler), the exception
246 will have the nodeid used for the old write enabler embedded in it,
247 in the following string::
249 The write enabler was recorded by nodeid '%s'.
251 Note that the nodeid here is encoded using the same base32 encoding
252 used by Foolscap and allmydata.util.idlib.nodeid_b2a().
255 return TupleOf(bool, DictOf(int, ReadData))
257 def advise_corrupt_share(share_type=str, storage_index=StorageIndex,
258 shnum=int, reason=str):
259 """Clients who discover hash failures in shares that they have
260 downloaded from me will use this method to inform me about the
261 failures. I will record their concern so that my operator can
262 manually inspect the shares in question. I return None.
264 'share_type' is either 'mutable' or 'immutable'. 'storage_index' is a
265 (binary) storage index string, and 'shnum' is the integer share
266 number. 'reason' is a human-readable explanation of the problem,
267 probably including some expected hash values and the computed ones
268 which did not match. Corruption advisories for mutable shares should
269 include a hash of the public key (the same value that appears in the
270 mutable-file verify-cap), since the current share format does not
274 class IStorageBucketWriter(Interface):
276 Objects of this kind live on the client side.
278 def put_block(segmentnum=int, data=ShareData):
279 """@param data: For most segments, this data will be 'blocksize'
280 bytes in length. The last segment might be shorter.
281 @return: a Deferred that fires (with None) when the operation completes
284 def put_plaintext_hashes(hashes=ListOf(Hash)):
286 @return: a Deferred that fires (with None) when the operation completes
289 def put_crypttext_hashes(hashes=ListOf(Hash)):
291 @return: a Deferred that fires (with None) when the operation completes
294 def put_block_hashes(blockhashes=ListOf(Hash)):
296 @return: a Deferred that fires (with None) when the operation completes
299 def put_share_hashes(sharehashes=ListOf(TupleOf(int, Hash))):
301 @return: a Deferred that fires (with None) when the operation completes
304 def put_uri_extension(data=URIExtensionData):
305 """This block of data contains integrity-checking information (hashes
306 of plaintext, crypttext, and shares), as well as encoding parameters
307 that are necessary to recover the data. This is a serialized dict
308 mapping strings to other strings. The hash of this data is kept in
309 the URI and verified before any of the data is used. All buckets for
310 a given file contain identical copies of this data.
312 The serialization format is specified with the following pseudocode:
313 for k in sorted(dict.keys()):
314 assert re.match(r'^[a-zA-Z_\-]+$', k)
315 write(k + ':' + netstring(dict[k]))
317 @return: a Deferred that fires (with None) when the operation completes
321 """Finish writing and close the bucket. The share is not finalized
322 until this method is called: if the uploading client disconnects
323 before calling close(), the partially-written share will be
326 @return: a Deferred that fires (with None) when the operation completes
329 class IStorageBucketReader(Interface):
331 def get_block_data(blocknum=int, blocksize=int, size=int):
332 """Most blocks will be the same size. The last block might be shorter
338 def get_crypttext_hashes():
340 @return: ListOf(Hash)
343 def get_block_hashes(at_least_these=SetOf(int)):
345 @return: ListOf(Hash)
348 def get_share_hashes(at_least_these=SetOf(int)):
350 @return: ListOf(TupleOf(int, Hash))
353 def get_uri_extension():
355 @return: URIExtensionData
358 class IStorageBroker(Interface):
359 def get_servers_for_psi(peer_selection_index):
361 @return: list of IServer instances
363 def get_connected_servers():
365 @return: frozenset of connected IServer instances
367 def get_known_servers():
369 @return: frozenset of IServer instances
371 def get_all_serverids():
373 @return: frozenset of serverid strings
375 def get_nickname_for_serverid(serverid):
377 @return: unicode nickname, or None
380 # methods moved from IntroducerClient, need review
381 def get_all_connections():
382 """Return a frozenset of (nodeid, service_name, rref) tuples, one for
383 each active connection we've established to a remote service. This is
384 mostly useful for unit tests that need to wait until a certain number
385 of connections have been made."""
387 def get_all_connectors():
388 """Return a dict that maps from (nodeid, service_name) to a
389 RemoteServiceConnector instance for all services that we are actively
390 trying to connect to. Each RemoteServiceConnector has the following
393 service_name: the type of service provided, like 'storage'
394 announcement_time: when we first heard about this service
395 last_connect_time: when we last established a connection
396 last_loss_time: when we last lost a connection
398 version: the peer's version, from the most recent connection
399 oldest_supported: the peer's oldest supported version, same
401 rref: the RemoteReference, if connected, otherwise None
402 remote_host: the IAddress, if connected, otherwise None
404 This method is intended for monitoring interfaces, such as a web page
405 which describes connecting and connected peers.
408 def get_all_peerids():
409 """Return a frozenset of all peerids to whom we have a connection (to
410 one or more services) established. Mostly useful for unit tests."""
412 def get_all_connections_for(service_name):
413 """Return a frozenset of (nodeid, service_name, rref) tuples, one
414 for each active connection that provides the given SERVICE_NAME."""
416 def get_permuted_peers(service_name, key):
417 """Returns an ordered list of (peerid, rref) tuples, selecting from
418 the connections that provide SERVICE_NAME, using a hash-based
419 permutation keyed by KEY. This randomizes the service list in a
420 repeatable way, to distribute load over many peers.
424 class IMutableSlotWriter(Interface):
426 The interface for a writer around a mutable slot on a remote server.
428 def set_checkstring(checkstring, *args):
430 Set the checkstring that I will pass to the remote server when
433 @param checkstring A packed checkstring to use.
435 Note that implementations can differ in which semantics they
436 wish to support for set_checkstring -- they can, for example,
437 build the checkstring themselves from its constituents, or
441 def get_checkstring():
443 Get the checkstring that I think currently exists on the remote
447 def put_block(data, segnum, salt):
449 Add a block and salt to the share.
452 def put_encprivey(encprivkey):
454 Add the encrypted private key to the share.
457 def put_blockhashes(blockhashes=list):
459 Add the block hash tree to the share.
462 def put_sharehashes(sharehashes=dict):
464 Add the share hash chain to the share.
469 Return the part of the share that needs to be signed.
472 def put_signature(signature):
474 Add the signature to the share.
477 def put_verification_key(verification_key):
479 Add the verification key to the share.
482 def finish_publishing():
484 Do anything necessary to finish writing the share to a remote
485 server. I require that no further publishing needs to take place
486 after this method has been called.
490 class IURI(Interface):
491 def init_from_string(uri):
492 """Accept a string (as created by my to_string() method) and populate
493 this instance with its data. I am not normally called directly,
494 please use the module-level uri.from_string() function to convert
495 arbitrary URI strings into IURI-providing instances."""
498 """Return False if this URI be used to modify the data. Return True
499 if this URI cannot be used to modify the data."""
502 """Return True if the data can be modified by *somebody* (perhaps
503 someone who has a more powerful URI than this one)."""
505 # TODO: rename to get_read_cap()
507 """Return another IURI instance, which represents a read-only form of
508 this one. If is_readonly() is True, this returns self."""
510 def get_verify_cap():
511 """Return an instance that provides IVerifierURI, which can be used
512 to check on the availability of the file or directory, without
513 providing enough capabilities to actually read or modify the
514 contents. This may return None if the file does not need checking or
515 verification (e.g. LIT URIs).
519 """Return a string of printable ASCII characters, suitable for
520 passing into init_from_string."""
522 class IVerifierURI(Interface, IURI):
523 def init_from_string(uri):
524 """Accept a string (as created by my to_string() method) and populate
525 this instance with its data. I am not normally called directly,
526 please use the module-level uri.from_string() function to convert
527 arbitrary URI strings into IURI-providing instances."""
530 """Return a string of printable ASCII characters, suitable for
531 passing into init_from_string."""
533 class IDirnodeURI(Interface):
534 """I am a URI which represents a dirnode."""
536 class IFileURI(Interface):
537 """I am a URI which represents a filenode."""
539 """Return the length (in bytes) of the file that I represent."""
541 class IImmutableFileURI(IFileURI):
544 class IMutableFileURI(Interface):
545 """I am a URI which represents a mutable filenode."""
546 def get_extension_params():
547 """Return the extension parameters in the URI"""
549 def set_extension_params():
550 """Set the extension parameters that should be in the URI"""
552 class IDirectoryURI(Interface):
555 class IReadonlyDirectoryURI(Interface):
558 class CapConstraintError(Exception):
559 """A constraint on a cap was violated."""
561 class MustBeDeepImmutableError(CapConstraintError):
562 """Mutable children cannot be added to an immutable directory.
563 Also, caps obtained from an immutable directory can trigger this error
564 if they are later found to refer to a mutable object and then used."""
566 class MustBeReadonlyError(CapConstraintError):
567 """Known write caps cannot be specified in a ro_uri field. Also,
568 caps obtained from a ro_uri field can trigger this error if they
569 are later found to be write caps and then used."""
571 class MustNotBeUnknownRWError(CapConstraintError):
572 """Cannot add an unknown child cap specified in a rw_uri field."""
575 class IReadable(Interface):
576 """I represent a readable object -- either an immutable file, or a
577 specific version of a mutable file.
581 """Return True if this reference provides mutable access to the given
582 file or directory (i.e. if you can modify it), or False if not. Note
583 that even if this reference is read-only, someone else may hold a
584 read-write reference to it.
586 For an IReadable returned by get_best_readable_version(), this will
587 always return True, but for instances of subinterfaces such as
588 IMutableFileVersion, it may return False."""
591 """Return True if this file or directory is mutable (by *somebody*,
592 not necessarily you), False if it is is immutable. Note that a file
593 might be mutable overall, but your reference to it might be
594 read-only. On the other hand, all references to an immutable file
595 will be read-only; there are no read-write references to an immutable
598 def get_storage_index():
599 """Return the storage index of the file."""
602 """Return the length (in bytes) of this readable object."""
604 def download_to_data():
605 """Download all of the file contents. I return a Deferred that fires
606 with the contents as a byte string."""
608 def read(consumer, offset=0, size=None):
609 """Download a portion (possibly all) of the file's contents, making
610 them available to the given IConsumer. Return a Deferred that fires
611 (with the consumer) when the consumer is unregistered (either because
612 the last byte has been given to it, or because the consumer threw an
613 exception during write(), possibly because it no longer wants to
614 receive data). The portion downloaded will start at 'offset' and
615 contain 'size' bytes (or the remainder of the file if size==None).
617 The consumer will be used in non-streaming mode: an IPullProducer
618 will be attached to it.
620 The consumer will not receive data right away: several network trips
621 must occur first. The order of events will be::
623 consumer.registerProducer(p, streaming)
624 (if streaming == False)::
625 consumer does p.resumeProducing()
627 consumer does p.resumeProducing()
628 consumer.write(data).. (repeat until all data is written)
629 consumer.unregisterProducer()
630 deferred.callback(consumer)
632 If a download error occurs, or an exception is raised by
633 consumer.registerProducer() or consumer.write(), I will call
634 consumer.unregisterProducer() and then deliver the exception via
635 deferred.errback(). To cancel the download, the consumer should call
636 p.stopProducing(), which will result in an exception being delivered
637 via deferred.errback().
639 See src/allmydata/util/consumer.py for an example of a simple
640 download-to-memory consumer.
644 class IWritable(Interface):
646 I define methods that callers can use to update SDMF and MDMF
647 mutable files on a Tahoe-LAFS grid.
649 # XXX: For the moment, we have only this. It is possible that we
650 # want to move overwrite() and modify() in here too.
651 def update(data, offset):
653 I write the data from my data argument to the MDMF file,
654 starting at offset. I continue writing data until my data
655 argument is exhausted, appending data to the file as necessary.
657 # assert IMutableUploadable.providedBy(data)
658 # to append data: offset=node.get_size_of_best_version()
659 # do we want to support compacting MDMF?
660 # for an MDMF file, this can be done with O(data.get_size())
661 # memory. For an SDMF file, any modification takes
662 # O(node.get_size_of_best_version()).
665 class IMutableFileVersion(IReadable):
666 """I provide access to a particular version of a mutable file. The
667 access is read/write if I was obtained from a filenode derived from
668 a write cap, or read-only if the filenode was derived from a read cap.
671 def get_sequence_number():
672 """Return the sequence number of this version."""
675 """Return the IMutableFileServerMap instance that was used to create
680 """Return this filenode's writekey, or None if the node does not have
681 write-capability. This may be used to assist with data structures
682 that need to make certain data available only to writers, such as the
683 read-write child caps in dirnodes. The recommended process is to have
684 reader-visible data be submitted to the filenode in the clear (where
685 it will be encrypted by the filenode using the readkey), but encrypt
686 writer-visible data using this writekey.
689 # TODO: Can this be overwrite instead of replace?
690 def replace(new_contents):
691 """Replace the contents of the mutable file, provided that no other
692 node has published (or is attempting to publish, concurrently) a
693 newer version of the file than this one.
695 I will avoid modifying any share that is different than the version
696 given by get_sequence_number(). However, if another node is writing
697 to the file at the same time as me, I may manage to update some shares
698 while they update others. If I see any evidence of this, I will signal
699 UncoordinatedWriteError, and the file will be left in an inconsistent
700 state (possibly the version you provided, possibly the old version,
701 possibly somebody else's version, and possibly a mix of shares from
704 The recommended response to UncoordinatedWriteError is to either
705 return it to the caller (since they failed to coordinate their
706 writes), or to attempt some sort of recovery. It may be sufficient to
707 wait a random interval (with exponential backoff) and repeat your
708 operation. If I do not signal UncoordinatedWriteError, then I was
709 able to write the new version without incident.
711 I return a Deferred that fires (with a PublishStatus object) when the
712 update has completed.
715 def modify(modifier_cb):
716 """Modify the contents of the file, by downloading this version,
717 applying the modifier function (or bound method), then uploading
718 the new version. This will succeed as long as no other node
719 publishes a version between the download and the upload.
720 I return a Deferred that fires (with a PublishStatus object) when
721 the update is complete.
723 The modifier callable will be given three arguments: a string (with
724 the old contents), a 'first_time' boolean, and a servermap. As with
725 download_to_data(), the old contents will be from this version,
726 but the modifier can use the servermap to make other decisions
727 (such as refusing to apply the delta if there are multiple parallel
728 versions, or if there is evidence of a newer unrecoverable version).
729 'first_time' will be True the first time the modifier is called,
730 and False on any subsequent calls.
732 The callable should return a string with the new contents. The
733 callable must be prepared to be called multiple times, and must
734 examine the input string to see if the change that it wants to make
735 is already present in the old version. If it does not need to make
736 any changes, it can either return None, or return its input string.
738 If the modifier raises an exception, it will be returned in the
743 # The hierarchy looks like this:
750 class IFilesystemNode(Interface):
752 """Return the strongest 'cap instance' associated with this node.
753 (writecap for writeable-mutable files/directories, readcap for
754 immutable or readonly-mutable files/directories). To convert this
755 into a string, call .to_string() on the result."""
758 """Return a readonly cap instance for this node. For immutable or
759 readonly nodes, get_cap() and get_readcap() return the same thing."""
761 def get_repair_cap():
762 """Return an IURI instance that can be used to repair the file, or
763 None if this node cannot be repaired (either because it is not
764 distributed, like a LIT file, or because the node does not represent
765 sufficient authority to create a repair-cap, like a read-only RSA
766 mutable file node [which cannot create the correct write-enablers]).
769 def get_verify_cap():
770 """Return an IVerifierURI instance that represents the
771 'verifiy/refresh capability' for this node. The holder of this
772 capability will be able to renew the lease for this node, protecting
773 it from garbage-collection. They will also be able to ask a server if
774 it holds a share for the file or directory.
778 """Return the URI string corresponding to the strongest cap associated
779 with this node. If this node is read-only, the URI will only offer
780 read-only access. If this node is read-write, the URI will offer
783 If you have read-write access to a node and wish to share merely
784 read-only access with others, use get_readonly_uri().
788 """Return the URI string that can be used by others to get write
789 access to this node, if it is writeable. If this is a read-only node,
792 def get_readonly_uri():
793 """Return the URI string that can be used by others to get read-only
794 access to this node. The result is a read-only URI, regardless of
795 whether this node is read-only or read-write.
797 If you have merely read-only access to this node, get_readonly_uri()
798 will return the same thing as get_uri().
801 def get_storage_index():
802 """Return a string with the (binary) storage index in use on this
803 download. This may be None if there is no storage index (i.e. LIT
804 files and directories)."""
807 """Return True if this reference provides mutable access to the given
808 file or directory (i.e. if you can modify it), or False if not. Note
809 that even if this reference is read-only, someone else may hold a
810 read-write reference to it."""
813 """Return True if this file or directory is mutable (by *somebody*,
814 not necessarily you), False if it is is immutable. Note that a file
815 might be mutable overall, but your reference to it might be
816 read-only. On the other hand, all references to an immutable file
817 will be read-only; there are no read-write references to an immutable
822 """Return True if this is an unknown node."""
824 def is_allowed_in_immutable_directory():
825 """Return True if this node is allowed as a child of a deep-immutable
826 directory. This is true if either the node is of a known-immutable type,
827 or it is unknown and read-only.
831 """Raise any error associated with this node."""
833 # XXX: These may not be appropriate outside the context of an IReadable.
835 """Return the length (in bytes) of the data this node represents. For
836 directory nodes, I return the size of the backing store. I return
837 synchronously and do not consult the network, so for mutable objects,
838 I will return the most recently observed size for the object, or None
839 if I don't remember a size. Use get_current_size, which returns a
840 Deferred, if you want more up-to-date information."""
842 def get_current_size():
843 """I return a Deferred that fires with the length (in bytes) of the
844 data this node represents.
847 class IFileNode(IFilesystemNode):
848 """I am a node which represents a file: a sequence of bytes. I am not a
849 container, like IDirectoryNode."""
850 def get_best_readable_version():
851 """Return a Deferred that fires with an IReadable for the 'best'
852 available version of the file. The IReadable provides only read
853 access, even if this filenode was derived from a write cap.
855 For an immutable file, there is only one version. For a mutable
856 file, the 'best' version is the recoverable version with the
857 highest sequence number. If no uncoordinated writes have occurred,
858 and if enough shares are available, then this will be the most
859 recent version that has been uploaded. If no version is recoverable,
860 the Deferred will errback with an UnrecoverableFileError.
863 def download_best_version():
864 """Download the contents of the version that would be returned
865 by get_best_readable_version(). This is equivalent to calling
866 download_to_data() on the IReadable given by that method.
868 I return a Deferred that fires with a byte string when the file
869 has been fully downloaded. To support streaming download, use
870 the 'read' method of IReadable. If no version is recoverable,
871 the Deferred will errback with an UnrecoverableFileError.
874 def get_size_of_best_version():
875 """Find the size of the version that would be returned by
876 get_best_readable_version().
878 I return a Deferred that fires with an integer. If no version
879 is recoverable, the Deferred will errback with an
880 UnrecoverableFileError.
884 class IImmutableFileNode(IFileNode, IReadable):
885 """I am a node representing an immutable file. Immutable files have
889 class IMutableFileNode(IFileNode):
890 """I provide access to a 'mutable file', which retains its identity
891 regardless of what contents are put in it.
893 The consistency-vs-availability problem means that there might be
894 multiple versions of a file present in the grid, some of which might be
895 unrecoverable (i.e. have fewer than 'k' shares). These versions are
896 loosely ordered: each has a sequence number and a hash, and any version
897 with seqnum=N was uploaded by a node which has seen at least one version
900 The 'servermap' (an instance of IMutableFileServerMap) is used to
901 describe the versions that are known to be present in the grid, and which
902 servers are hosting their shares. It is used to represent the 'state of
903 the world', and is used for this purpose by my test-and-set operations.
904 Downloading the contents of the mutable file will also return a
905 servermap. Uploading a new version into the mutable file requires a
906 servermap as input, and the semantics of the replace operation is
907 'replace the file with my new version if it looks like nobody else has
908 changed the file since my previous download'. Because the file is
909 distributed, this is not a perfect test-and-set operation, but it will do
910 its best. If the replace process sees evidence of a simultaneous write,
911 it will signal an UncoordinatedWriteError, so that the caller can take
915 Most readers will want to use the 'best' current version of the file, and
916 should use my 'download_best_version()' method.
918 To unconditionally replace the file, callers should use overwrite(). This
919 is the mode that user-visible mutable files will probably use.
921 To apply some delta to the file, call modify() with a callable modifier
922 function that can apply the modification that you want to make. This is
923 the mode that dirnodes will use, since most directory modification
924 operations can be expressed in terms of deltas to the directory state.
927 Three methods are available for users who need to perform more complex
928 operations. The first is get_servermap(), which returns an up-to-date
929 servermap using a specified mode. The second is download_version(), which
930 downloads a specific version (not necessarily the 'best' one). The third
931 is 'upload', which accepts new contents and a servermap (which must have
932 been updated with MODE_WRITE). The upload method will attempt to apply
933 the new contents as long as no other node has modified the file since the
934 servermap was updated. This might be useful to a caller who wants to
935 merge multiple versions into a single new one.
937 Note that each time the servermap is updated, a specific 'mode' is used,
938 which determines how many peers are queried. To use a servermap for my
939 replace() method, that servermap must have been updated in MODE_WRITE.
940 These modes are defined in allmydata.mutable.common, and consist of
941 MODE_READ, MODE_WRITE, MODE_ANYTHING, and MODE_CHECK. Please look in
942 allmydata/mutable/servermap.py for details about the differences.
944 Mutable files are currently limited in size (about 3.5MB max) and can
945 only be retrieved and updated all-at-once, as a single big string. Future
946 versions of our mutable files will remove this restriction.
948 def get_best_mutable_version():
949 """Return a Deferred that fires with an IMutableFileVersion for
950 the 'best' available version of the file. The best version is
951 the recoverable version with the highest sequence number. If no
952 uncoordinated writes have occurred, and if enough shares are
953 available, then this will be the most recent version that has
956 If no version is recoverable, the Deferred will errback with an
957 UnrecoverableFileError.
960 def overwrite(new_contents):
961 """Unconditionally replace the contents of the mutable file with new
962 ones. This simply chains get_servermap(MODE_WRITE) and upload(). This
963 is only appropriate to use when the new contents of the file are
964 completely unrelated to the old ones, and you do not care about other
967 I return a Deferred that fires (with a PublishStatus object) when the
968 update has completed.
971 def modify(modifier_cb):
972 """Modify the contents of the file, by downloading the current
973 version, applying the modifier function (or bound method), then
974 uploading the new version. I return a Deferred that fires (with a
975 PublishStatus object) when the update is complete.
977 The modifier callable will be given three arguments: a string (with
978 the old contents), a 'first_time' boolean, and a servermap. As with
979 download_best_version(), the old contents will be from the best
980 recoverable version, but the modifier can use the servermap to make
981 other decisions (such as refusing to apply the delta if there are
982 multiple parallel versions, or if there is evidence of a newer
983 unrecoverable version). 'first_time' will be True the first time the
984 modifier is called, and False on any subsequent calls.
986 The callable should return a string with the new contents. The
987 callable must be prepared to be called multiple times, and must
988 examine the input string to see if the change that it wants to make
989 is already present in the old version. If it does not need to make
990 any changes, it can either return None, or return its input string.
992 If the modifier raises an exception, it will be returned in the
996 def get_servermap(mode):
997 """Return a Deferred that fires with an IMutableFileServerMap
998 instance, updated using the given mode.
1001 def download_version(servermap, version):
1002 """Download a specific version of the file, using the servermap
1003 as a guide to where the shares are located.
1005 I return a Deferred that fires with the requested contents, or
1006 errbacks with UnrecoverableFileError. Note that a servermap which was
1007 updated with MODE_ANYTHING or MODE_READ may not know about shares for
1008 all versions (those modes stop querying servers as soon as they can
1009 fulfil their goals), so you may want to use MODE_CHECK (which checks
1010 everything) to get increased visibility.
1013 def upload(new_contents, servermap):
1014 """Replace the contents of the file with new ones. This requires a
1015 servermap that was previously updated with MODE_WRITE.
1017 I attempt to provide test-and-set semantics, in that I will avoid
1018 modifying any share that is different than the version I saw in the
1019 servermap. However, if another node is writing to the file at the
1020 same time as me, I may manage to update some shares while they update
1021 others. If I see any evidence of this, I will signal
1022 UncoordinatedWriteError, and the file will be left in an inconsistent
1023 state (possibly the version you provided, possibly the old version,
1024 possibly somebody else's version, and possibly a mix of shares from
1027 The recommended response to UncoordinatedWriteError is to either
1028 return it to the caller (since they failed to coordinate their
1029 writes), or to attempt some sort of recovery. It may be sufficient to
1030 wait a random interval (with exponential backoff) and repeat your
1031 operation. If I do not signal UncoordinatedWriteError, then I was
1032 able to write the new version without incident.
1034 I return a Deferred that fires (with a PublishStatus object) when the
1035 publish has completed. I will update the servermap in-place with the
1036 location of all new shares.
1040 """Return this filenode's writekey, or None if the node does not have
1041 write-capability. This may be used to assist with data structures
1042 that need to make certain data available only to writers, such as the
1043 read-write child caps in dirnodes. The recommended process is to have
1044 reader-visible data be submitted to the filenode in the clear (where
1045 it will be encrypted by the filenode using the readkey), but encrypt
1046 writer-visible data using this writekey.
1050 """Returns the mutable file protocol version."""
1052 class NotEnoughSharesError(Exception):
1053 """Download was unable to get enough shares"""
1055 class NoSharesError(Exception):
1056 """Download was unable to get any shares at all."""
1058 class UploadUnhappinessError(Exception):
1059 """Upload was unable to satisfy 'servers_of_happiness'"""
1061 class UnableToFetchCriticalDownloadDataError(Exception):
1062 """I was unable to fetch some piece of critical data which is supposed to
1063 be identically present in all shares."""
1065 class NoServersError(Exception):
1066 """Upload wasn't given any servers to work with, usually indicating a
1067 network or Introducer problem."""
1069 class ExistingChildError(Exception):
1070 """A directory node was asked to add or replace a child that already
1071 exists, and overwrite= was set to False."""
1073 class NoSuchChildError(Exception):
1074 """A directory node was asked to fetch a child which does not exist."""
1076 class ChildOfWrongTypeError(Exception):
1077 """An operation was attempted on a child of the wrong type (file or directory)."""
1079 class IDirectoryNode(IFilesystemNode):
1080 """I represent a filesystem node that is a container, with a
1081 name-to-child mapping, holding the tahoe equivalent of a directory. All
1082 child names are unicode strings, and all children are some sort of
1083 IFilesystemNode (a file, subdirectory, or unknown node).
1088 The dirnode ('1') URI returned by this method can be used in
1089 set_uri() on a different directory ('2') to 'mount' a reference to
1090 this directory ('1') under the other ('2'). This URI is just a
1091 string, so it can be passed around through email or other out-of-band
1095 def get_readonly_uri():
1097 The dirnode ('1') URI returned by this method can be used in
1098 set_uri() on a different directory ('2') to 'mount' a reference to
1099 this directory ('1') under the other ('2'). This URI is just a
1100 string, so it can be passed around through email or other out-of-band
1105 """I return a Deferred that fires with a dictionary mapping child
1106 name (a unicode string) to (node, metadata_dict) tuples, in which
1107 'node' is an IFilesystemNode and 'metadata_dict' is a dictionary of
1110 def has_child(name):
1111 """I return a Deferred that fires with a boolean, True if there
1112 exists a child of the given name, False if not. The child name must
1113 be a unicode string."""
1116 """I return a Deferred that fires with a specific named child node,
1117 which is an IFilesystemNode. The child name must be a unicode string.
1118 I raise NoSuchChildError if I do not have a child by that name."""
1120 def get_metadata_for(name):
1121 """I return a Deferred that fires with the metadata dictionary for
1122 a specific named child node. The child name must be a unicode string.
1123 This metadata is stored in the *edge*, not in the child, so it is
1124 attached to the parent dirnode rather than the child node.
1125 I raise NoSuchChildError if I do not have a child by that name."""
1127 def set_metadata_for(name, metadata):
1128 """I replace any existing metadata for the named child with the new
1129 metadata. The child name must be a unicode string. This metadata is
1130 stored in the *edge*, not in the child, so it is attached to the
1131 parent dirnode rather than the child node. I return a Deferred
1132 (that fires with this dirnode) when the operation is complete.
1133 I raise NoSuchChildError if I do not have a child by that name."""
1135 def get_child_at_path(path):
1136 """Transform a child path into an IFilesystemNode.
1138 I perform a recursive series of 'get' operations to find the named
1139 descendant node. I return a Deferred that fires with the node, or
1140 errbacks with NoSuchChildError if the node could not be found.
1142 The path can be either a single string (slash-separated) or a list of
1143 path-name elements. All elements must be unicode strings.
1146 def get_child_and_metadata_at_path(path):
1147 """Transform a child path into an IFilesystemNode and metadata.
1149 I am like get_child_at_path(), but my Deferred fires with a tuple of
1150 (node, metadata). The metadata comes from the last edge. If the path
1151 is empty, the metadata will be an empty dictionary.
1154 def set_uri(name, writecap, readcap=None, metadata=None, overwrite=True):
1155 """I add a child (by writecap+readcap) at the specific name. I return
1156 a Deferred that fires when the operation finishes. If overwrite= is
1157 True, I will replace any existing child of the same name, otherwise
1158 an existing child will cause me to return ExistingChildError. The
1159 child name must be a unicode string.
1161 The child caps could be for a file, or for a directory. If you have
1162 both the writecap and readcap, you should provide both arguments.
1163 If you have only one cap and don't know whether it is read-only,
1164 provide it as the writecap argument and leave the readcap as None.
1165 If you have only one cap that is known to be read-only, provide it
1166 as the readcap argument and leave the writecap as None.
1167 The filecaps are typically obtained from an IFilesystemNode with
1168 get_uri() and get_readonly_uri().
1170 If metadata= is provided, I will use it as the metadata for the named
1171 edge. This will replace any existing metadata. If metadata= is left
1172 as the default value of None, I will set ['mtime'] to the current
1173 time, and I will set ['ctime'] to the current time if there was not
1174 already a child by this name present. This roughly matches the
1175 ctime/mtime semantics of traditional filesystems. See the
1176 "About the metadata" section of webapi.txt for futher information.
1178 If this directory node is read-only, the Deferred will errback with a
1179 NotWriteableError."""
1181 def set_children(entries, overwrite=True):
1182 """Add multiple children (by writecap+readcap) to a directory node.
1183 Takes a dictionary, with childname as keys and (writecap, readcap)
1184 tuples (or (writecap, readcap, metadata) triples) as values. Returns
1185 a Deferred that fires (with this dirnode) when the operation
1186 finishes. This is equivalent to calling set_uri() multiple times, but
1187 is much more efficient. All child names must be unicode strings.
1190 def set_node(name, child, metadata=None, overwrite=True):
1191 """I add a child at the specific name. I return a Deferred that fires
1192 when the operation finishes. This Deferred will fire with the child
1193 node that was just added. I will replace any existing child of the
1194 same name. The child name must be a unicode string. The 'child'
1195 instance must be an instance providing IFilesystemNode.
1197 If metadata= is provided, I will use it as the metadata for the named
1198 edge. This will replace any existing metadata. If metadata= is left
1199 as the default value of None, I will set ['mtime'] to the current
1200 time, and I will set ['ctime'] to the current time if there was not
1201 already a child by this name present. This roughly matches the
1202 ctime/mtime semantics of traditional filesystems. See the
1203 "About the metadata" section of webapi.txt for futher information.
1205 If this directory node is read-only, the Deferred will errback with a
1206 NotWriteableError."""
1208 def set_nodes(entries, overwrite=True):
1209 """Add multiple children to a directory node. Takes a dict mapping
1210 unicode childname to (child_node, metdata) tuples. If metdata=None,
1211 the original metadata is left unmodified. Returns a Deferred that
1212 fires (with this dirnode) when the operation finishes. This is
1213 equivalent to calling set_node() multiple times, but is much more
1216 def add_file(name, uploadable, metadata=None, overwrite=True):
1217 """I upload a file (using the given IUploadable), then attach the
1218 resulting ImmutableFileNode to the directory at the given name. I set
1219 metadata the same way as set_uri and set_node. The child name must be
1222 I return a Deferred that fires (with the IFileNode of the uploaded
1223 file) when the operation completes."""
1225 def delete(name, must_exist=True, must_be_directory=False, must_be_file=False):
1226 """I remove the child at the specific name. I return a Deferred that
1227 fires when the operation finishes. The child name must be a unicode
1228 string. If must_exist is True and I do not have a child by that name,
1229 I raise NoSuchChildError. If must_be_directory is True and the child
1230 is a file, or if must_be_file is True and the child is a directory,
1231 I raise ChildOfWrongTypeError."""
1233 def create_subdirectory(name, initial_children={}, overwrite=True, metadata=None):
1234 """I create and attach a directory at the given name. The new
1235 directory can be empty, or it can be populated with children
1236 according to 'initial_children', which takes a dictionary in the same
1237 format as set_nodes (i.e. mapping unicode child name to (childnode,
1238 metadata) tuples). The child name must be a unicode string. I return
1239 a Deferred that fires (with the new directory node) when the
1240 operation finishes."""
1242 def move_child_to(current_child_name, new_parent, new_child_name=None,
1244 """I take one of my children and move them to a new parent. The child
1245 is referenced by name. On the new parent, the child will live under
1246 'new_child_name', which defaults to 'current_child_name'. TODO: what
1247 should we do about metadata? I return a Deferred that fires when the
1248 operation finishes. The child name must be a unicode string. I raise
1249 NoSuchChildError if I do not have a child by that name."""
1251 def build_manifest():
1252 """I generate a table of everything reachable from this directory.
1253 I also compute deep-stats as described below.
1255 I return a Monitor. The Monitor's results will be a dictionary with
1258 res['manifest']: a list of (path, cap) tuples for all nodes
1259 (directories and files) reachable from this one.
1260 'path' will be a tuple of unicode strings. The
1261 origin dirnode will be represented by an empty path
1263 res['verifycaps']: a list of (printable) verifycap strings, one for
1264 each reachable non-LIT node. This is a set:
1265 it will contain no duplicates.
1266 res['storage-index']: a list of (base32) storage index strings,
1267 one for each reachable non-LIT node. This is
1268 a set: it will contain no duplicates.
1269 res['stats']: a dictionary, the same that is generated by
1270 start_deep_stats() below.
1272 The Monitor will also have an .origin_si attribute with the (binary)
1273 storage index of the starting point.
1276 def start_deep_stats():
1277 """Return a Monitor, examining all nodes (directories and files)
1278 reachable from this one. The Monitor's results will be a dictionary
1279 with the following keys::
1281 count-immutable-files: count of how many CHK files are in the set
1282 count-mutable-files: same, for mutable files (does not include
1284 count-literal-files: same, for LIT files
1285 count-files: sum of the above three
1287 count-directories: count of directories
1289 size-immutable-files: total bytes for all CHK files in the set
1290 size-mutable-files (TODO): same, for current version of all mutable
1291 files, does not include directories
1292 size-literal-files: same, for LIT files
1293 size-directories: size of mutable files used by directories
1295 largest-directory: number of bytes in the largest directory
1296 largest-directory-children: number of children in the largest
1298 largest-immutable-file: number of bytes in the largest CHK file
1300 size-mutable-files is not yet implemented, because it would involve
1301 even more queries than deep_stats does.
1303 The Monitor will also have an .origin_si attribute with the (binary)
1304 storage index of the starting point.
1306 This operation will visit every directory node underneath this one,
1307 and can take a long time to run. On a typical workstation with good
1308 bandwidth, this can examine roughly 15 directories per second (and
1309 takes several minutes of 100% CPU for ~1700 directories).
1312 class ICodecEncoder(Interface):
1313 def set_params(data_size, required_shares, max_shares):
1314 """Set up the parameters of this encoder.
1316 This prepares the encoder to perform an operation that converts a
1317 single block of data into a number of shares, such that a future
1318 ICodecDecoder can use a subset of these shares to recover the
1319 original data. This operation is invoked by calling encode(). Once
1320 the encoding parameters are set up, the encode operation can be
1321 invoked multiple times.
1323 set_params() prepares the encoder to accept blocks of input data that
1324 are exactly 'data_size' bytes in length. The encoder will be prepared
1325 to produce 'max_shares' shares for each encode() operation (although
1326 see the 'desired_share_ids' to use less CPU). The encoding math will
1327 be chosen such that the decoder can get by with as few as
1328 'required_shares' of these shares and still reproduce the original
1329 data. For example, set_params(1000, 5, 5) offers no redundancy at
1330 all, whereas set_params(1000, 1, 10) provides 10x redundancy.
1332 Numerical Restrictions: 'data_size' is required to be an integral
1333 multiple of 'required_shares'. In general, the caller should choose
1334 required_shares and max_shares based upon their reliability
1335 requirements and the number of peers available (the total storage
1336 space used is roughly equal to max_shares*data_size/required_shares),
1337 then choose data_size to achieve the memory footprint desired (larger
1338 data_size means more efficient operation, smaller data_size means
1339 smaller memory footprint).
1341 In addition, 'max_shares' must be equal to or greater than
1342 'required_shares'. Of course, setting them to be equal causes
1343 encode() to degenerate into a particularly slow form of the 'split'
1346 See encode() for more details about how these parameters are used.
1348 set_params() must be called before any other ICodecEncoder methods
1353 """Return the 3-tuple of data_size, required_shares, max_shares"""
1355 def get_encoder_type():
1356 """Return a short string that describes the type of this encoder.
1358 There is required to be a global table of encoder classes. This method
1359 returns an index into this table; the value at this index is an
1360 encoder class, and this encoder is an instance of that class.
1363 def get_block_size():
1364 """Return the length of the shares that encode() will produce.
1367 def encode_proposal(data, desired_share_ids=None):
1368 """Encode some data.
1370 'data' must be a string (or other buffer object), and len(data) must
1371 be equal to the 'data_size' value passed earlier to set_params().
1373 This will return a Deferred that will fire with two lists. The first
1374 is a list of shares, each of which is a string (or other buffer
1375 object) such that len(share) is the same as what get_share_size()
1376 returned earlier. The second is a list of shareids, in which each is
1377 an integer. The lengths of the two lists will always be equal to each
1378 other. The user should take care to keep each share closely
1379 associated with its shareid, as one is useless without the other.
1381 The length of this output list will normally be the same as the value
1382 provided to the 'max_shares' parameter of set_params(). This may be
1383 different if 'desired_share_ids' is provided.
1385 'desired_share_ids', if provided, is required to be a sequence of
1386 ints, each of which is required to be >= 0 and < max_shares. If not
1387 provided, encode() will produce 'max_shares' shares, as if
1388 'desired_share_ids' were set to range(max_shares). You might use this
1389 if you initially thought you were going to use 10 peers, started
1390 encoding, and then two of the peers dropped out: you could use
1391 desired_share_ids= to skip the work (both memory and CPU) of
1392 producing shares for the peers which are no longer available.
1396 def encode(inshares, desired_share_ids=None):
1397 """Encode some data. This may be called multiple times. Each call is
1400 inshares is a sequence of length required_shares, containing buffers
1401 (i.e. strings), where each buffer contains the next contiguous
1402 non-overlapping segment of the input data. Each buffer is required to
1403 be the same length, and the sum of the lengths of the buffers is
1404 required to be exactly the data_size promised by set_params(). (This
1405 implies that the data has to be padded before being passed to
1406 encode(), unless of course it already happens to be an even multiple
1407 of required_shares in length.)
1409 Note: the requirement to break up your data into
1410 'required_shares' chunks of exactly the right length before
1411 calling encode() is surprising from point of view of a user
1412 who doesn't know how FEC works. It feels like an
1413 implementation detail that has leaked outside the abstraction
1414 barrier. Is there a use case in which the data to be encoded
1415 might already be available in pre-segmented chunks, such that
1416 it is faster or less work to make encode() take a list rather
1417 than splitting a single string?
1419 Yes, there is: suppose you are uploading a file with K=64,
1420 N=128, segsize=262,144. Then each in-share will be of size
1421 4096. If you use this .encode() API then your code could first
1422 read each successive 4096-byte chunk from the file and store
1423 each one in a Python string and store each such Python string
1424 in a Python list. Then you could call .encode(), passing that
1425 list as "inshares". The encoder would generate the other 64
1426 "secondary shares" and return to you a new list containing
1427 references to the same 64 Python strings that you passed in
1428 (as the primary shares) plus references to the new 64 Python
1431 (You could even imagine that your code could use readv() so
1432 that the operating system can arrange to get all of those
1433 bytes copied from the file into the Python list of Python
1434 strings as efficiently as possible instead of having a loop
1435 written in C or in Python to copy the next part of the file
1436 into the next string.)
1438 On the other hand if you instead use the .encode_proposal()
1439 API (above), then your code can first read in all of the
1440 262,144 bytes of the segment from the file into a Python
1441 string, then call .encode_proposal() passing the segment data
1442 as the "data" argument. The encoder would basically first
1443 split the "data" argument into a list of 64 in-shares of 4096
1444 byte each, and then do the same thing that .encode() does. So
1445 this would result in a little bit more copying of data and a
1446 little bit higher of a "maximum memory usage" during the
1447 process, although it might or might not make a practical
1448 difference for our current use cases.
1450 Note that "inshares" is a strange name for the parameter if
1451 you think of the parameter as being just for feeding in data
1452 to the codec. It makes more sense if you think of the result
1453 of this encoding as being the set of shares from inshares plus
1454 an extra set of "secondary shares" (or "check shares"). It is
1455 a surprising name! If the API is going to be surprising then
1456 the name should be surprising. If we switch to
1457 encode_proposal() above then we should also switch to an
1460 'desired_share_ids', if provided, is required to be a sequence of
1461 ints, each of which is required to be >= 0 and < max_shares. If not
1462 provided, encode() will produce 'max_shares' shares, as if
1463 'desired_share_ids' were set to range(max_shares). You might use this
1464 if you initially thought you were going to use 10 peers, started
1465 encoding, and then two of the peers dropped out: you could use
1466 desired_share_ids= to skip the work (both memory and CPU) of
1467 producing shares for the peers which are no longer available.
1469 For each call, encode() will return a Deferred that fires with two
1470 lists, one containing shares and the other containing the shareids.
1471 The get_share_size() method can be used to determine the length of
1472 the share strings returned by encode(). Each shareid is a small
1473 integer, exactly as passed into 'desired_share_ids' (or
1474 range(max_shares), if desired_share_ids was not provided).
1476 The shares and their corresponding shareids are required to be kept
1477 together during storage and retrieval. Specifically, the share data is
1478 useless by itself: the decoder needs to be told which share is which
1479 by providing it with both the shareid and the actual share data.
1481 This function will allocate an amount of memory roughly equal to::
1483 (max_shares - required_shares) * get_share_size()
1485 When combined with the memory that the caller must allocate to
1486 provide the input data, this leads to a memory footprint roughly
1487 equal to the size of the resulting encoded shares (i.e. the expansion
1488 factor times the size of the input segment).
1493 # returning a list of (shareidN,shareN) tuples instead of a pair of
1494 # lists (shareids..,shares..). Brian thought the tuples would
1495 # encourage users to keep the share and shareid together throughout
1496 # later processing, Zooko pointed out that the code to iterate
1497 # through two lists is not really more complicated than using a list
1498 # of tuples and there's also a performance improvement
1500 # having 'data_size' not required to be an integral multiple of
1501 # 'required_shares'. Doing this would require encode() to perform
1502 # padding internally, and we'd prefer to have any padding be done
1503 # explicitly by the caller. Yes, it is an abstraction leak, but
1504 # hopefully not an onerous one.
1507 class ICodecDecoder(Interface):
1508 def set_params(data_size, required_shares, max_shares):
1509 """Set the params. They have to be exactly the same ones that were
1510 used for encoding."""
1512 def get_needed_shares():
1513 """Return the number of shares needed to reconstruct the data.
1514 set_params() is required to be called before this."""
1516 def decode(some_shares, their_shareids):
1517 """Decode a partial list of shares into data.
1519 'some_shares' is required to be a sequence of buffers of sharedata, a
1520 subset of the shares returned by ICodecEncode.encode(). Each share is
1521 required to be of the same length. The i'th element of their_shareids
1522 is required to be the shareid of the i'th buffer in some_shares.
1524 This returns a Deferred which fires with a sequence of buffers. This
1525 sequence will contain all of the segments of the original data, in
1526 order. The sum of the lengths of all of the buffers will be the
1527 'data_size' value passed into the original ICodecEncode.set_params()
1528 call. To get back the single original input block of data, use
1529 ''.join(output_buffers), or you may wish to simply write them in
1530 order to an output file.
1532 Note that some of the elements in the result sequence may be
1533 references to the elements of the some_shares input sequence. In
1534 particular, this means that if those share objects are mutable (e.g.
1535 arrays) and if they are changed, then both the input (the
1536 'some_shares' parameter) and the output (the value given when the
1537 deferred is triggered) will change.
1539 The length of 'some_shares' is required to be exactly the value of
1540 'required_shares' passed into the original ICodecEncode.set_params()
1544 class IEncoder(Interface):
1545 """I take an object that provides IEncryptedUploadable, which provides
1546 encrypted data, and a list of shareholders. I then encode, hash, and
1547 deliver shares to those shareholders. I will compute all the necessary
1548 Merkle hash trees that are necessary to validate the crypttext that
1549 eventually comes back from the shareholders. I provide the URI Extension
1550 Block Hash, and the encoding parameters, both of which must be included
1553 I do not choose shareholders, that is left to the IUploader. I must be
1554 given a dict of RemoteReferences to storage buckets that are ready and
1555 willing to receive data.
1559 """Specify the number of bytes that will be encoded. This must be
1560 peformed before get_serialized_params() can be called.
1562 def set_params(params):
1563 """Override the default encoding parameters. 'params' is a tuple of
1564 (k,d,n), where 'k' is the number of required shares, 'd' is the
1565 servers_of_happiness, and 'n' is the total number of shares that will
1568 Encoding parameters can be set in three ways. 1: The Encoder class
1569 provides defaults (3/7/10). 2: the Encoder can be constructed with
1570 an 'options' dictionary, in which the
1571 needed_and_happy_and_total_shares' key can be a (k,d,n) tuple. 3:
1572 set_params((k,d,n)) can be called.
1574 If you intend to use set_params(), you must call it before
1575 get_share_size or get_param are called.
1578 def set_encrypted_uploadable(u):
1579 """Provide a source of encrypted upload data. 'u' must implement
1580 IEncryptedUploadable.
1582 When this is called, the IEncryptedUploadable will be queried for its
1583 length and the storage_index that should be used.
1585 This returns a Deferred that fires with this Encoder instance.
1587 This must be performed before start() can be called.
1590 def get_param(name):
1591 """Return an encoding parameter, by name.
1593 'storage_index': return a string with the (16-byte truncated SHA-256
1594 hash) storage index to which these shares should be
1597 'share_counts': return a tuple describing how many shares are used:
1598 (needed_shares, servers_of_happiness, total_shares)
1600 'num_segments': return an int with the number of segments that
1603 'segment_size': return an int with the size of each segment.
1605 'block_size': return the size of the individual blocks that will
1606 be delivered to a shareholder's put_block() method. By
1607 knowing this, the shareholder will be able to keep all
1608 blocks in a single file and still provide random access
1609 when reading them. # TODO: can we avoid exposing this?
1611 'share_size': an int with the size of the data that will be stored
1612 on each shareholder. This is aggregate amount of data
1613 that will be sent to the shareholder, summed over all
1614 the put_block() calls I will ever make. It is useful to
1615 determine this size before asking potential
1616 shareholders whether they will grant a lease or not,
1617 since their answers will depend upon how much space we
1618 need. TODO: this might also include some amount of
1619 overhead, like the size of all the hashes. We need to
1620 decide whether this is useful or not.
1622 'serialized_params': a string with a concise description of the
1623 codec name and its parameters. This may be passed
1624 into the IUploadable to let it make sure that
1625 the same file encoded with different parameters
1626 will result in different storage indexes.
1628 Once this is called, set_size() and set_params() may not be called.
1631 def set_shareholders(shareholders, servermap):
1632 """Tell the encoder where to put the encoded shares. 'shareholders'
1633 must be a dictionary that maps share number (an integer ranging from
1634 0 to n-1) to an instance that provides IStorageBucketWriter.
1635 'servermap' is a dictionary that maps share number (as defined above)
1636 to a set of peerids. This must be performed before start() can be
1640 """Begin the encode/upload process. This involves reading encrypted
1641 data from the IEncryptedUploadable, encoding it, uploading the shares
1642 to the shareholders, then sending the hash trees.
1644 set_encrypted_uploadable() and set_shareholders() must be called
1645 before this can be invoked.
1647 This returns a Deferred that fires with a verify cap when the upload
1648 process is complete. The verifycap, plus the encryption key, is
1649 sufficient to construct the read cap.
1652 class IDecoder(Interface):
1653 """I take a list of shareholders and some setup information, then
1654 download, validate, decode, and decrypt data from them, writing the
1655 results to an output file.
1657 I do not locate the shareholders, that is left to the IDownloader. I must
1658 be given a dict of RemoteReferences to storage buckets that are ready to
1663 """I take a file-like object (providing write and close) to which all
1664 the plaintext data will be written.
1666 TODO: producer/consumer . Maybe write() should return a Deferred that
1667 indicates when it will accept more data? But probably having the
1668 IDecoder be a producer is easier to glue to IConsumer pieces.
1671 def set_shareholders(shareholders):
1672 """I take a dictionary that maps share identifiers (small integers)
1673 to RemoteReferences that provide RIBucketReader. This must be called
1677 """I start the download. This process involves retrieving data and
1678 hash chains from the shareholders, using the hashes to validate the
1679 data, decoding the shares into segments, decrypting the segments,
1680 then writing the resulting plaintext to the output file.
1682 I return a Deferred that will fire (with self) when the download is
1686 class IDownloadTarget(Interface):
1687 # Note that if the IDownloadTarget is also an IConsumer, the downloader
1688 # will register itself as a producer. This allows the target to invoke
1689 # downloader.pauseProducing, resumeProducing, and stopProducing.
1691 """Called before any calls to write() or close(). If an error
1692 occurs before any data is available, fail() may be called without
1693 a previous call to open().
1695 'size' is the length of the file being downloaded, in bytes."""
1698 """Output some data to the target."""
1700 """Inform the target that there is no more data to be written."""
1702 """fail() is called to indicate that the download has failed. 'why'
1703 is a Failure object indicating what went wrong. No further methods
1704 will be invoked on the IDownloadTarget after fail()."""
1705 def register_canceller(cb):
1706 """The CiphertextDownloader uses this to register a no-argument function
1707 that the target can call to cancel the download. Once this canceller
1708 is invoked, no further calls to write() or close() will be made."""
1710 """When the CiphertextDownloader is done, this finish() function will be
1711 called. Whatever it returns will be returned to the invoker of
1712 Downloader.download.
1715 class IDownloader(Interface):
1716 def download(uri, target):
1717 """Perform a CHK download, sending the data to the given target.
1718 'target' must provide IDownloadTarget.
1720 Returns a Deferred that fires (with the results of target.finish)
1721 when the download is finished, or errbacks if something went wrong."""
1723 class IEncryptedUploadable(Interface):
1724 def set_upload_status(upload_status):
1725 """Provide an IUploadStatus object that should be filled with status
1726 information. The IEncryptedUploadable is responsible for setting
1727 key-determination progress ('chk'), size, storage_index, and
1728 ciphertext-fetch progress. It may delegate some of this
1729 responsibility to others, in particular to the IUploadable."""
1732 """This behaves just like IUploadable.get_size()."""
1734 def get_all_encoding_parameters():
1735 """Return a Deferred that fires with a tuple of
1736 (k,happy,n,segment_size). The segment_size will be used as-is, and
1737 must match the following constraints: it must be a multiple of k, and
1738 it shouldn't be unreasonably larger than the file size (if
1739 segment_size is larger than filesize, the difference must be stored
1742 This usually passes through to the IUploadable method of the same
1745 The encoder strictly obeys the values returned by this method. To
1746 make an upload use non-default encoding parameters, you must arrange
1747 to control the values that this method returns.
1750 def get_storage_index():
1751 """Return a Deferred that fires with a 16-byte storage index.
1754 def read_encrypted(length, hash_only):
1755 """This behaves just like IUploadable.read(), but returns crypttext
1756 instead of plaintext. If hash_only is True, then this discards the
1757 data (and returns an empty list); this improves efficiency when
1758 resuming an interrupted upload (where we need to compute the
1759 plaintext hashes, but don't need the redundant encrypted data)."""
1761 def get_plaintext_hashtree_leaves(first, last, num_segments):
1762 """OBSOLETE; Get the leaf nodes of a merkle hash tree over the
1763 plaintext segments, i.e. get the tagged hashes of the given segments.
1764 The segment size is expected to be generated by the
1765 IEncryptedUploadable before any plaintext is read or ciphertext
1766 produced, so that the segment hashes can be generated with only a
1769 This returns a Deferred which fires with a sequence of hashes, using:
1771 tuple(segment_hashes[first:last])
1773 'num_segments' is used to assert that the number of segments that the
1774 IEncryptedUploadable handled matches the number of segments that the
1775 encoder was expecting.
1777 This method must not be called until the final byte has been read
1778 from read_encrypted(). Once this method is called, read_encrypted()
1779 can never be called again.
1782 def get_plaintext_hash():
1783 """OBSOLETE; Get the hash of the whole plaintext.
1785 This returns a Deferred which fires with a tagged SHA-256 hash of the
1786 whole plaintext, obtained from hashutil.plaintext_hash(data).
1790 """Just like IUploadable.close()."""
1792 class IUploadable(Interface):
1793 def set_upload_status(upload_status):
1794 """Provide an IUploadStatus object that should be filled with status
1795 information. The IUploadable is responsible for setting
1796 key-determination progress ('chk')."""
1798 def set_default_encoding_parameters(params):
1799 """Set the default encoding parameters, which must be a dict mapping
1800 strings to ints. The meaningful keys are 'k', 'happy', 'n', and
1801 'max_segment_size'. These might have an influence on the final
1802 encoding parameters returned by get_all_encoding_parameters(), if the
1803 Uploadable doesn't have more specific preferences.
1805 This call is optional: if it is not used, the Uploadable will use
1806 some built-in defaults. If used, this method must be called before
1807 any other IUploadable methods to have any effect.
1811 """Return a Deferred that will fire with the length of the data to be
1812 uploaded, in bytes. This will be called before the data is actually
1813 used, to compute encoding parameters.
1816 def get_all_encoding_parameters():
1817 """Return a Deferred that fires with a tuple of
1818 (k,happy,n,segment_size). The segment_size will be used as-is, and
1819 must match the following constraints: it must be a multiple of k, and
1820 it shouldn't be unreasonably larger than the file size (if
1821 segment_size is larger than filesize, the difference must be stored
1824 The relative values of k and n allow some IUploadables to request
1825 better redundancy than others (in exchange for consuming more space
1828 Larger values of segment_size reduce hash overhead, while smaller
1829 values reduce memory footprint and cause data to be delivered in
1830 smaller pieces (which may provide a smoother and more predictable
1831 download experience).
1833 The encoder strictly obeys the values returned by this method. To
1834 make an upload use non-default encoding parameters, you must arrange
1835 to control the values that this method returns. One way to influence
1836 them may be to call set_encoding_parameters() before calling
1837 get_all_encoding_parameters().
1840 def get_encryption_key():
1841 """Return a Deferred that fires with a 16-byte AES key. This key will
1842 be used to encrypt the data. The key will also be hashed to derive
1845 Uploadables which want to achieve convergence should hash their file
1846 contents and the serialized_encoding_parameters to form the key
1847 (which of course requires a full pass over the data). Uploadables can
1848 use the upload.ConvergentUploadMixin class to achieve this
1851 Uploadables which do not care about convergence (or do not wish to
1852 make multiple passes over the data) can simply return a
1853 strongly-random 16 byte string.
1855 get_encryption_key() may be called multiple times: the IUploadable is
1856 required to return the same value each time.
1860 """Return a Deferred that fires with a list of strings (perhaps with
1861 only a single element) which, when concatenated together, contain the
1862 next 'length' bytes of data. If EOF is near, this may provide fewer
1863 than 'length' bytes. The total number of bytes provided by read()
1864 before it signals EOF must equal the size provided by get_size().
1866 If the data must be acquired through multiple internal read
1867 operations, returning a list instead of a single string may help to
1868 reduce string copies. However, the length of the concatenated strings
1869 must equal the amount of data requested, unless EOF is encountered.
1870 Long reads, or short reads without EOF, are not allowed. read()
1871 should return the same amount of data as a local disk file read, just
1872 in a different shape and asynchronously.
1874 'length' will typically be equal to (min(get_size(),1MB)/req_shares),
1875 so a 10kB file means length=3kB, 100kB file means length=30kB,
1876 and >=1MB file means length=300kB.
1878 This method provides for a single full pass through the data. Later
1879 use cases may desire multiple passes or access to only parts of the
1880 data (such as a mutable file making small edits-in-place). This API
1881 will be expanded once those use cases are better understood.
1885 """The upload is finished, and whatever filehandle was in use may be
1889 class IMutableUploadable(Interface):
1891 I represent content that is due to be uploaded to a mutable filecap.
1893 # This is somewhat simpler than the IUploadable interface above
1894 # because mutable files do not need to be concerned with possibly
1895 # generating a CHK, nor with per-file keys. It is a subset of the
1896 # methods in IUploadable, though, so we could just as well implement
1897 # the mutable uploadables as IUploadables that don't happen to use
1898 # those methods (with the understanding that the unused methods will
1899 # never be called on such objects)
1902 Returns a Deferred that fires with the size of the content held
1908 Returns a list of strings which, when concatenated, are the next
1909 length bytes of the file, or fewer if there are fewer bytes
1910 between the current location and the end of the file.
1915 The process that used the Uploadable is finished using it, so
1916 the uploadable may be closed.
1919 class IUploadResults(Interface):
1920 """I am returned by upload() methods. I contain a number of public
1921 attributes which can be read to determine the results of the upload. Some
1922 of these are functional, some are timing information. All of these may be
1925 .file_size : the size of the file, in bytes
1926 .uri : the CHK read-cap for the file
1927 .ciphertext_fetched : how many bytes were fetched by the helper
1928 .sharemap: dict mapping share identifier to set of serverids
1929 (binary strings). This indicates which servers were given
1930 which shares. For immutable files, the shareid is an
1931 integer (the share number, from 0 to N-1). For mutable
1932 files, it is a string of the form 'seq%d-%s-sh%d',
1933 containing the sequence number, the roothash, and the
1935 .servermap : dict mapping server peerid to a set of share numbers
1936 .timings : dict of timing information, mapping name to seconds (float)
1937 total : total upload time, start to finish
1938 storage_index : time to compute the storage index
1939 peer_selection : time to decide which peers will be used
1940 contacting_helper : initial helper query to upload/no-upload decision
1941 existence_check : helper pre-upload existence check
1942 helper_total : initial helper query to helper finished pushing
1943 cumulative_fetch : helper waiting for ciphertext requests
1944 total_fetch : helper start to last ciphertext response
1945 cumulative_encoding : just time spent in zfec
1946 cumulative_sending : just time spent waiting for storage servers
1947 hashes_and_close : last segment push to shareholder close
1948 total_encode_and_push : first encode to shareholder close
1952 class IDownloadResults(Interface):
1953 """I am created internally by download() methods. I contain a number of
1954 public attributes which contain details about the download process.::
1956 .file_size : the size of the file, in bytes
1957 .servers_used : set of server peerids that were used during download
1958 .server_problems : dict mapping server peerid to a problem string. Only
1959 servers that had problems (bad hashes, disconnects)
1961 .servermap : dict mapping server peerid to a set of share numbers. Only
1962 servers that had any shares are listed here.
1963 .timings : dict of timing information, mapping name to seconds (float)
1964 peer_selection : time to ask servers about shares
1965 servers_peer_selection : dict of peerid to DYHB-query time
1966 uri_extension : time to fetch a copy of the URI extension block
1967 hashtrees : time to fetch the hash trees
1968 segments : time to fetch, decode, and deliver segments
1969 cumulative_fetch : time spent waiting for storage servers
1970 cumulative_decode : just time spent in zfec
1971 cumulative_decrypt : just time spent in decryption
1972 total : total download time, start to finish
1973 fetch_per_server : dict of peerid to list of per-segment fetch times
1977 class IUploader(Interface):
1978 def upload(uploadable):
1979 """Upload the file. 'uploadable' must impement IUploadable. This
1980 returns a Deferred which fires with an IUploadResults instance, from
1981 which the URI of the file can be obtained as results.uri ."""
1983 def upload_ssk(write_capability, new_version, uploadable):
1984 """TODO: how should this work?"""
1986 class ICheckable(Interface):
1987 def check(monitor, verify=False, add_lease=False):
1988 """Check up on my health, optionally repairing any problems.
1990 This returns a Deferred that fires with an instance that provides
1991 ICheckResults, or None if the object is non-distributed (i.e. LIT
1994 The monitor will be checked periodically to see if the operation has
1995 been cancelled. If so, no new queries will be sent, and the Deferred
1996 will fire (with a OperationCancelledError) immediately.
1998 Filenodes and dirnodes (which provide IFilesystemNode) are also
1999 checkable. Instances that represent verifier-caps will be checkable
2000 but not downloadable. Some objects (like LIT files) do not actually
2001 live in the grid, and their checkers return None (non-distributed
2002 files are always healthy).
2004 If verify=False, a relatively lightweight check will be performed: I
2005 will ask all servers if they have a share for me, and I will believe
2006 whatever they say. If there are at least N distinct shares on the
2007 grid, my results will indicate r.is_healthy()==True. This requires a
2008 roundtrip to each server, but does not transfer very much data, so
2009 the network bandwidth is fairly low.
2011 If verify=True, a more resource-intensive check will be performed:
2012 every share will be downloaded, and the hashes will be validated on
2013 every bit. I will ignore any shares that failed their hash checks. If
2014 there are at least N distinct valid shares on the grid, my results
2015 will indicate r.is_healthy()==True. This requires N/k times as much
2016 download bandwidth (and server disk IO) as a regular download. If a
2017 storage server is holding a corrupt share, or is experiencing memory
2018 failures during retrieval, or is malicious or buggy, then
2019 verification will detect the problem, but checking will not.
2021 If add_lease=True, I will ensure that an up-to-date lease is present
2022 on each share. The lease secrets will be derived from by node secret
2023 (in BASEDIR/private/secret), so either I will add a new lease to the
2024 share, or I will merely renew the lease that I already had. In a
2025 future version of the storage-server protocol (once Accounting has
2026 been implemented), there may be additional options here to define the
2027 kind of lease that is obtained (which account number to claim, etc).
2029 TODO: any problems seen during checking will be reported to the
2030 health-manager.furl, a centralized object which is responsible for
2031 figuring out why files are unhealthy so corrective action can be
2035 def check_and_repair(monitor, verify=False, add_lease=False):
2036 """Like check(), but if the file/directory is not healthy, attempt to
2039 Any non-healthy result will cause an immediate repair operation, to
2040 generate and upload new shares. After repair, the file will be as
2041 healthy as we can make it. Details about what sort of repair is done
2042 will be put in the check-and-repair results. The Deferred will not
2043 fire until the repair is complete.
2045 This returns a Deferred which fires with an instance of
2046 ICheckAndRepairResults."""
2048 class IDeepCheckable(Interface):
2049 def start_deep_check(verify=False, add_lease=False):
2050 """Check upon the health of me and everything I can reach.
2052 This is a recursive form of check(), useable only on dirnodes.
2054 I return a Monitor, with results that are an IDeepCheckResults
2057 TODO: If any of the directories I traverse are unrecoverable, the
2058 Monitor will report failure. If any of the files I check upon are
2059 unrecoverable, those problems will be reported in the
2060 IDeepCheckResults as usual, and the Monitor will not report a
2064 def start_deep_check_and_repair(verify=False, add_lease=False):
2065 """Check upon the health of me and everything I can reach. Repair
2066 anything that isn't healthy.
2068 This is a recursive form of check_and_repair(), useable only on
2071 I return a Monitor, with results that are an
2072 IDeepCheckAndRepairResults object.
2074 TODO: If any of the directories I traverse are unrecoverable, the
2075 Monitor will report failure. If any of the files I check upon are
2076 unrecoverable, those problems will be reported in the
2077 IDeepCheckResults as usual, and the Monitor will not report a
2081 class ICheckResults(Interface):
2082 """I contain the detailed results of a check/verify operation.
2085 def get_storage_index():
2086 """Return a string with the (binary) storage index."""
2087 def get_storage_index_string():
2088 """Return a string with the (printable) abbreviated storage index."""
2090 """Return the (string) URI of the object that was checked."""
2093 """Return a boolean, True if the file/dir is fully healthy, False if
2094 it is damaged in any way. Non-distributed LIT files always return
2097 def is_recoverable():
2098 """Return a boolean, True if the file/dir can be recovered, False if
2099 not. Unrecoverable files are obviously unhealthy. Non-distributed LIT
2100 files always return True."""
2102 def needs_rebalancing():
2103 """Return a boolean, True if the file/dir's reliability could be
2104 improved by moving shares to new servers. Non-distributed LIT files
2105 always return False."""
2109 """Return a dictionary that describes the state of the file/dir. LIT
2110 files always return an empty dictionary. Normal files and directories
2111 return a dictionary with the following keys (note that these use
2112 binary strings rather than base32-encoded ones) (also note that for
2113 mutable files, these counts are for the 'best' version):
2115 count-shares-good: the number of distinct good shares that were found
2116 count-shares-needed: 'k', the number of shares required for recovery
2117 count-shares-expected: 'N', the number of total shares generated
2118 count-good-share-hosts: the number of distinct storage servers with
2119 good shares. If this number is less than
2120 count-shares-good, then some shares are
2121 doubled up, increasing the correlation of
2122 failures. This indicates that one or more
2123 shares should be moved to an otherwise unused
2124 server, if one is available.
2125 count-corrupt-shares: the number of shares with integrity failures
2126 list-corrupt-shares: a list of 'share locators', one for each share
2127 that was found to be corrupt. Each share
2128 locator is a list of (serverid, storage_index,
2130 count-incompatible-shares: the number of shares which are of a share
2131 format unknown to this checker
2132 list-incompatible-shares: a list of 'share locators', one for each
2133 share that was found to be of an unknown
2134 format. Each share locator is a list of
2135 (serverid, storage_index, sharenum).
2136 servers-responding: list of (binary) storage server identifiers,
2137 one for each server which responded to the share
2138 query (even if they said they didn't have
2139 shares, and even if they said they did have
2140 shares but then didn't send them when asked, or
2141 dropped the connection, or returned a Failure,
2142 and even if they said they did have shares and
2143 sent incorrect ones when asked)
2144 sharemap: dict mapping share identifier to list of serverids
2145 (binary strings). This indicates which servers are holding
2146 which shares. For immutable files, the shareid is an
2147 integer (the share number, from 0 to N-1). For mutable
2148 files, it is a string of the form 'seq%d-%s-sh%d',
2149 containing the sequence number, the roothash, and the
2152 The following keys are most relevant for mutable files, but immutable
2153 files will provide sensible values too::
2155 count-wrong-shares: the number of shares for versions other than the
2156 'best' one (which is defined as being the
2157 recoverable version with the highest sequence
2158 number, then the highest roothash). These are
2159 either leftover shares from an older version
2160 (perhaps on a server that was offline when an
2161 update occurred), shares from an unrecoverable
2162 newer version, or shares from an alternate
2163 current version that results from an
2164 uncoordinated write collision. For a healthy
2165 file, this will equal 0.
2167 count-recoverable-versions: the number of recoverable versions of
2168 the file. For a healthy file, this will
2171 count-unrecoverable-versions: the number of unrecoverable versions
2172 of the file. For a healthy file, this
2178 """Return a string with a brief (one-line) summary of the results."""
2181 """Return a list of strings with more detailed results."""
2183 class ICheckAndRepairResults(Interface):
2184 """I contain the detailed results of a check/verify/repair operation.
2186 The IFilesystemNode.check()/verify()/repair() methods all return
2187 instances that provide ICheckAndRepairResults.
2190 def get_storage_index():
2191 """Return a string with the (binary) storage index."""
2192 def get_storage_index_string():
2193 """Return a string with the (printable) abbreviated storage index."""
2194 def get_repair_attempted():
2195 """Return a boolean, True if a repair was attempted. We might not
2196 attempt to repair the file because it was healthy, or healthy enough
2197 (i.e. some shares were missing but not enough to exceed some
2198 threshold), or because we don't know how to repair this object."""
2199 def get_repair_successful():
2200 """Return a boolean, True if repair was attempted and the file/dir
2201 was fully healthy afterwards. False if no repair was attempted or if
2202 a repair attempt failed."""
2203 def get_pre_repair_results():
2204 """Return an ICheckResults instance that describes the state of the
2205 file/dir before any repair was attempted."""
2206 def get_post_repair_results():
2207 """Return an ICheckResults instance that describes the state of the
2208 file/dir after any repair was attempted. If no repair was attempted,
2209 the pre-repair and post-repair results will be identical."""
2212 class IDeepCheckResults(Interface):
2213 """I contain the results of a deep-check operation.
2215 This is returned by a call to ICheckable.deep_check().
2218 def get_root_storage_index_string():
2219 """Return the storage index (abbreviated human-readable string) of
2220 the first object checked."""
2222 """Return a dictionary with the following keys::
2224 count-objects-checked: count of how many objects were checked
2225 count-objects-healthy: how many of those objects were completely
2227 count-objects-unhealthy: how many were damaged in some way
2228 count-objects-unrecoverable: how many were unrecoverable
2229 count-corrupt-shares: how many shares were found to have
2230 corruption, summed over all objects
2234 def get_corrupt_shares():
2235 """Return a set of (serverid, storage_index, sharenum) for all shares
2236 that were found to be corrupt. Both serverid and storage_index are
2239 def get_all_results():
2240 """Return a dictionary mapping pathname (a tuple of strings, ready to
2241 be slash-joined) to an ICheckResults instance, one for each object
2242 that was checked."""
2244 def get_results_for_storage_index(storage_index):
2245 """Retrive the ICheckResults instance for the given (binary)
2246 storage index. Raises KeyError if there are no results for that
2250 """Return a dictionary with the same keys as
2251 IDirectoryNode.deep_stats()."""
2253 class IDeepCheckAndRepairResults(Interface):
2254 """I contain the results of a deep-check-and-repair operation.
2256 This is returned by a call to ICheckable.deep_check_and_repair().
2259 def get_root_storage_index_string():
2260 """Return the storage index (abbreviated human-readable string) of
2261 the first object checked."""
2263 """Return a dictionary with the following keys::
2265 count-objects-checked: count of how many objects were checked
2266 count-objects-healthy-pre-repair: how many of those objects were
2267 completely healthy (before any
2269 count-objects-unhealthy-pre-repair: how many were damaged in
2271 count-objects-unrecoverable-pre-repair: how many were unrecoverable
2272 count-objects-healthy-post-repair: how many of those objects were
2273 completely healthy (after any
2275 count-objects-unhealthy-post-repair: how many were damaged in
2277 count-objects-unrecoverable-post-repair: how many were
2279 count-repairs-attempted: repairs were attempted on this many
2280 objects. The count-repairs- keys will
2281 always be provided, however unless
2282 repair=true is present, they will all
2284 count-repairs-successful: how many repairs resulted in healthy
2286 count-repairs-unsuccessful: how many repairs resulted did not
2287 results in completely healthy objects
2288 count-corrupt-shares-pre-repair: how many shares were found to
2289 have corruption, summed over all
2290 objects examined (before any
2292 count-corrupt-shares-post-repair: how many shares were found to
2293 have corruption, summed over all
2294 objects examined (after any
2299 """Return a dictionary with the same keys as
2300 IDirectoryNode.deep_stats()."""
2302 def get_corrupt_shares():
2303 """Return a set of (serverid, storage_index, sharenum) for all shares
2304 that were found to be corrupt before any repair was attempted. Both
2305 serverid and storage_index are binary.
2307 def get_remaining_corrupt_shares():
2308 """Return a set of (serverid, storage_index, sharenum) for all shares
2309 that were found to be corrupt after any repair was completed. Both
2310 serverid and storage_index are binary. These are shares that need
2311 manual inspection and probably deletion.
2313 def get_all_results():
2314 """Return a dictionary mapping pathname (a tuple of strings, ready to
2315 be slash-joined) to an ICheckAndRepairResults instance, one for each
2316 object that was checked."""
2318 def get_results_for_storage_index(storage_index):
2319 """Retrive the ICheckAndRepairResults instance for the given (binary)
2320 storage index. Raises KeyError if there are no results for that
2324 class IRepairable(Interface):
2325 def repair(check_results):
2326 """Attempt to repair the given object. Returns a Deferred that fires
2327 with a IRepairResults object.
2329 I must be called with an object that implements ICheckResults, as
2330 proof that you have actually discovered a problem with this file. I
2331 will use the data in the checker results to guide the repair process,
2332 such as which servers provided bad data and should therefore be
2333 avoided. The ICheckResults object is inside the
2334 ICheckAndRepairResults object, which is returned by the
2335 ICheckable.check() method::
2337 d = filenode.check(repair=False)
2338 def _got_results(check_and_repair_results):
2339 check_results = check_and_repair_results.get_pre_repair_results()
2340 return filenode.repair(check_results)
2341 d.addCallback(_got_results)
2345 class IRepairResults(Interface):
2346 """I contain the results of a repair operation."""
2347 def get_successful(self):
2348 """Returns a boolean: True if the repair made the file healthy, False
2349 if not. Repair failure generally indicates a file that has been
2350 damaged beyond repair."""
2353 class IClient(Interface):
2354 def upload(uploadable):
2355 """Upload some data into a CHK, get back the UploadResults for it.
2356 @param uploadable: something that implements IUploadable
2357 @return: a Deferred that fires with the UploadResults instance.
2358 To get the URI for this file, use results.uri .
2361 def create_mutable_file(contents=""):
2362 """Create a new mutable file (with initial) contents, get back the
2365 @param contents: (bytestring, callable, or None): this provides the
2366 initial contents of the mutable file. If 'contents' is a bytestring,
2367 it will be used as-is. If 'contents' is a callable, it will be
2368 invoked with the new MutableFileNode instance and is expected to
2369 return a bytestring with the initial contents of the file (the
2370 callable can use node.get_writekey() to decide how to encrypt the
2371 initial contents, e.g. for a brand new dirnode with initial
2372 children). contents=None is equivalent to an empty string. Using
2373 content_maker= is more efficient than creating a mutable file and
2374 setting its contents in two separate operations.
2376 @return: a Deferred that fires with an IMutableFileNode instance.
2379 def create_dirnode(initial_children={}):
2380 """Create a new unattached dirnode, possibly with initial children.
2382 @param initial_children: dict with keys that are unicode child names,
2383 and values that are (childnode, metadata) tuples.
2385 @return: a Deferred that fires with the new IDirectoryNode instance.
2388 def create_node_from_uri(uri, rouri):
2389 """Create a new IFilesystemNode instance from the uri, synchronously.
2390 @param uri: a string or IURI-providing instance, or None. This could
2391 be for a LiteralFileNode, a CHK file node, a mutable file
2392 node, or a directory node
2393 @param rouri: a string or IURI-providing instance, or None. If the
2394 main uri is None, I will use the rouri instead. If I
2395 recognize the format of the main uri, I will ignore the
2396 rouri (because it can be derived from the writecap).
2398 @return: an instance that provides IFilesystemNode (or more usefully
2399 one of its subclasses). File-specifying URIs will result in
2400 IFileNode-providing instances, like ImmutableFileNode,
2401 LiteralFileNode, or MutableFileNode. Directory-specifying
2402 URIs will result in IDirectoryNode-providing instances, like
2406 class INodeMaker(Interface):
2407 """The NodeMaker is used to create IFilesystemNode instances. It can
2408 accept a filecap/dircap string and return the node right away. It can
2409 also create new nodes (i.e. upload a file, or create a mutable file)
2410 asynchronously. Once you have one of these nodes, you can use other
2411 methods to determine whether it is a file or directory, and to download
2412 or modify its contents.
2414 The NodeMaker encapsulates all the authorities that these
2415 IFilesystemNodes require (like references to the StorageFarmBroker). Each
2416 Tahoe process will typically have a single NodeMaker, but unit tests may
2417 create simplified/mocked forms for testing purposes.
2419 def create_from_cap(writecap, readcap=None, **kwargs):
2420 """I create an IFilesystemNode from the given writecap/readcap. I can
2421 only provide nodes for existing file/directory objects: use my other
2422 methods to create new objects. I return synchronously."""
2424 def create_mutable_file(contents=None, keysize=None):
2425 """I create a new mutable file, and return a Deferred which will fire
2426 with the IMutableFileNode instance when it is ready. If contents= is
2427 provided (a bytestring), it will be used as the initial contents of
2428 the new file, otherwise the file will contain zero bytes. keysize= is
2429 for use by unit tests, to create mutable files that are smaller than
2432 def create_new_mutable_directory(initial_children={}):
2433 """I create a new mutable directory, and return a Deferred which will
2434 fire with the IDirectoryNode instance when it is ready. If
2435 initial_children= is provided (a dict mapping unicode child name to
2436 (childnode, metadata_dict) tuples), the directory will be populated
2437 with those children, otherwise it will be empty."""
2439 class IClientStatus(Interface):
2440 def list_all_uploads():
2441 """Return a list of uploader objects, one for each upload which
2442 currently has an object available (tracked with weakrefs). This is
2443 intended for debugging purposes."""
2444 def list_active_uploads():
2445 """Return a list of active IUploadStatus objects."""
2446 def list_recent_uploads():
2447 """Return a list of IUploadStatus objects for the most recently
2450 def list_all_downloads():
2451 """Return a list of downloader objects, one for each download which
2452 currently has an object available (tracked with weakrefs). This is
2453 intended for debugging purposes."""
2454 def list_active_downloads():
2455 """Return a list of active IDownloadStatus objects."""
2456 def list_recent_downloads():
2457 """Return a list of IDownloadStatus objects for the most recently
2458 started downloads."""
2460 class IUploadStatus(Interface):
2462 """Return a timestamp (float with seconds since epoch) indicating
2463 when the operation was started."""
2464 def get_storage_index():
2465 """Return a string with the (binary) storage index in use on this
2466 upload. Returns None if the storage index has not yet been
2469 """Return an integer with the number of bytes that will eventually
2470 be uploaded for this file. Returns None if the size is not yet known.
2473 """Return True if this upload is using a Helper, False if not."""
2475 """Return a string describing the current state of the upload
2478 """Returns a tuple of floats, (chk, ciphertext, encode_and_push),
2479 each from 0.0 to 1.0 . 'chk' describes how much progress has been
2480 made towards hashing the file to determine a CHK encryption key: if
2481 non-convergent encryption is in use, this will be trivial, otherwise
2482 the whole file must be hashed. 'ciphertext' describes how much of the
2483 ciphertext has been pushed to the helper, and is '1.0' for non-helper
2484 uploads. 'encode_and_push' describes how much of the encode-and-push
2485 process has finished: for helper uploads this is dependent upon the
2486 helper providing progress reports. It might be reasonable to add all
2487 three numbers and report the sum to the user."""
2489 """Return True if the upload is currently active, False if not."""
2491 """Return an instance of UploadResults (which contains timing and
2492 sharemap information). Might return None if the upload is not yet
2495 """Each upload status gets a unique number: this method returns that
2496 number. This provides a handle to this particular upload, so a web
2497 page can generate a suitable hyperlink."""
2499 class IDownloadStatus(Interface):
2501 """Return a timestamp (float with seconds since epoch) indicating
2502 when the operation was started."""
2503 def get_storage_index():
2504 """Return a string with the (binary) storage index in use on this
2505 download. This may be None if there is no storage index (i.e. LIT
2508 """Return an integer with the number of bytes that will eventually be
2509 retrieved for this file. Returns None if the size is not yet known.
2512 """Return True if this download is using a Helper, False if not."""
2514 """Return a string describing the current state of the download
2517 """Returns a float (from 0.0 to 1.0) describing the amount of the
2518 download that has completed. This value will remain at 0.0 until the
2519 first byte of plaintext is pushed to the download target."""
2521 """Return True if the download is currently active, False if not."""
2523 """Each download status gets a unique number: this method returns
2524 that number. This provides a handle to this particular download, so a
2525 web page can generate a suitable hyperlink."""
2527 class IServermapUpdaterStatus(Interface):
2529 class IPublishStatus(Interface):
2531 class IRetrieveStatus(Interface):
2534 class NotCapableError(Exception):
2535 """You have tried to write to a read-only node."""
2537 class BadWriteEnablerError(Exception):
2540 class RIControlClient(RemoteInterface):
2542 def wait_for_client_connections(num_clients=int):
2543 """Do not return until we have connections to at least NUM_CLIENTS
2547 def upload_from_file_to_uri(filename=str,
2548 convergence=ChoiceOf(None,
2549 StringConstraint(2**20))):
2550 """Upload a file to the grid. This accepts a filename (which must be
2551 absolute) that points to a file on the node's local disk. The node will
2552 read the contents of this file, upload it to the grid, then return the
2553 URI at which it was uploaded. If convergence is None then a random
2554 encryption key will be used, else the plaintext will be hashed, then
2555 that hash will be mixed together with the "convergence" string to form
2560 def download_from_uri_to_file(uri=URI, filename=str):
2561 """Download a file from the grid, placing it on the node's local disk
2562 at the given filename (which must be absolute[?]). Returns the
2563 absolute filename where the file was written."""
2568 def get_memory_usage():
2569 """Return a dict describes the amount of memory currently in use. The
2570 keys are 'VmPeak', 'VmSize', and 'VmData'. The values are integers,
2571 measuring memory consupmtion in bytes."""
2572 return DictOf(str, int)
2574 def speed_test(count=int, size=int, mutable=Any()):
2575 """Write 'count' tempfiles to disk, all of the given size. Measure
2576 how long (in seconds) it takes to upload them all to the servers.
2577 Then measure how long it takes to download all of them. If 'mutable'
2578 is 'create', time creation of mutable files. If 'mutable' is
2579 'upload', then time access to the same mutable file instead of
2582 Returns a tuple of (upload_time, download_time).
2584 return (float, float)
2586 def measure_peer_response_time():
2587 """Send a short message to each connected peer, and measure the time
2588 it takes for them to respond to it. This is a rough measure of the
2589 application-level round trip time.
2591 @return: a dictionary mapping peerid to a float (RTT time in seconds)
2594 return DictOf(str, float)
2596 UploadResults = Any() #DictOf(str, str)
2598 class RIEncryptedUploadable(RemoteInterface):
2599 __remote_name__ = "RIEncryptedUploadable.tahoe.allmydata.com"
2604 def get_all_encoding_parameters():
2605 return (int, int, int, long)
2607 def read_encrypted(offset=Offset, length=ReadSize):
2614 class RICHKUploadHelper(RemoteInterface):
2615 __remote_name__ = "RIUploadHelper.tahoe.allmydata.com"
2619 Return a dictionary of version information.
2621 return DictOf(str, Any())
2623 def upload(reader=RIEncryptedUploadable):
2624 return UploadResults
2627 class RIHelper(RemoteInterface):
2628 __remote_name__ = "RIHelper.tahoe.allmydata.com"
2632 Return a dictionary of version information.
2634 return DictOf(str, Any())
2636 def upload_chk(si=StorageIndex):
2637 """See if a file with a given storage index needs uploading. The
2638 helper will ask the appropriate storage servers to see if the file
2639 has already been uploaded. If so, the helper will return a set of
2640 'upload results' that includes whatever hashes are needed to build
2641 the read-cap, and perhaps a truncated sharemap.
2643 If the file has not yet been uploaded (or if it was only partially
2644 uploaded), the helper will return an empty upload-results dictionary
2645 and also an RICHKUploadHelper object that will take care of the
2646 upload process. The client should call upload() on this object and
2647 pass it a reference to an RIEncryptedUploadable object that will
2648 provide ciphertext. When the upload is finished, the upload() method
2649 will finish and return the upload results.
2651 return (UploadResults, ChoiceOf(RICHKUploadHelper, None))
2654 class RIStatsProvider(RemoteInterface):
2655 __remote_name__ = "RIStatsProvider.tahoe.allmydata.com"
2657 Provides access to statistics and monitoring information.
2662 returns a dictionary containing 'counters' and 'stats', each a
2663 dictionary with string counter/stat name keys, and numeric or None values.
2664 counters are monotonically increasing measures of work done, and
2665 stats are instantaneous measures (potentially time averaged
2668 return DictOf(str, DictOf(str, ChoiceOf(float, int, long, None)))
2670 class RIStatsGatherer(RemoteInterface):
2671 __remote_name__ = "RIStatsGatherer.tahoe.allmydata.com"
2673 Provides a monitoring service for centralised collection of stats
2676 def provide(provider=RIStatsProvider, nickname=str):
2678 @param provider: a stats collector instance which should be polled
2679 periodically by the gatherer to collect stats.
2680 @param nickname: a name useful to identify the provided client
2685 class IStatsProducer(Interface):
2688 returns a dictionary, with str keys representing the names of stats
2689 to be monitored, and numeric values.
2692 class RIKeyGenerator(RemoteInterface):
2693 __remote_name__ = "RIKeyGenerator.tahoe.allmydata.com"
2695 Provides a service offering to make RSA key pairs.
2698 def get_rsa_key_pair(key_size=int):
2700 @param key_size: the size of the signature key.
2701 @return: tuple(verifying_key, signing_key)
2703 return TupleOf(str, str)
2706 class FileTooLargeError(Exception):
2709 class IValidatedThingProxy(Interface):
2711 """ Acquire a thing and validate it. Return a deferred which is
2712 eventually fired with self if the thing is valid or errbacked if it
2713 can't be acquired or validated."""
2715 class InsufficientVersionError(Exception):
2716 def __init__(self, needed, got):
2717 self.needed = needed
2720 return "InsufficientVersionError(need '%s', got %s)" % (self.needed,
2723 class EmptyPathnameComponentError(Exception):
2724 """The webapi disallows empty pathname components."""