mirror of
https://github.com/markqvist/Reticulum.git
synced 2024-11-20 04:31:34 +00:00
1090 lines
45 KiB
Python
1090 lines
45 KiB
Python
# MIT License
|
|
#
|
|
# Copyright (c) 2016-2022 Mark Qvist / unsigned.io
|
|
#
|
|
# Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
# of this software and associated documentation files (the "Software"), to deal
|
|
# in the Software without restriction, including without limitation the rights
|
|
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
# copies of the Software, and to permit persons to whom the Software is
|
|
# furnished to do so, subject to the following conditions:
|
|
#
|
|
# The above copyright notice and this permission notice shall be included in all
|
|
# copies or substantial portions of the Software.
|
|
#
|
|
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
# SOFTWARE.
|
|
|
|
import RNS
|
|
import os
|
|
import bz2
|
|
import math
|
|
import time
|
|
import threading
|
|
from .vendor import umsgpack as umsgpack
|
|
from time import sleep
|
|
|
|
class Resource:
|
|
"""
|
|
The Resource class allows transferring arbitrary amounts
|
|
of data over a link. It will automatically handle sequencing,
|
|
compression, coordination and checksumming.
|
|
|
|
:param data: The data to be transferred. Can be *bytes* or an open *file handle*. See the :ref:`Filetransfer Example<example-filetransfer>` for details.
|
|
:param link: The :ref:`RNS.Link<api-link>` instance on which to transfer the data.
|
|
:param advertise: Optional. Whether to automatically advertise the resource. Can be *True* or *False*.
|
|
:param auto_compress: Optional. Whether to auto-compress the resource. Can be *True* or *False*.
|
|
:param callback: An optional *callable* with the signature *callback(resource)*. Will be called when the resource transfer concludes.
|
|
:param progress_callback: An optional *callable* with the signature *callback(resource)*. Will be called whenever the resource transfer progress is updated.
|
|
"""
|
|
|
|
# The initial window size at beginning of transfer
|
|
WINDOW = 4
|
|
|
|
# Absolute minimum window size during transfer
|
|
WINDOW_MIN = 1
|
|
|
|
# The maximum window size for transfers on slow links
|
|
WINDOW_MAX_SLOW = 10
|
|
|
|
# The maximum window size for transfers on fast links
|
|
WINDOW_MAX_FAST = 75
|
|
|
|
# For calculating maps and guard segments, this
|
|
# must be set to the global maximum window.
|
|
WINDOW_MAX = WINDOW_MAX_FAST
|
|
|
|
# If the fast rate is sustained for this many request
|
|
# rounds, the fast link window size will be allowed.
|
|
FAST_RATE_THRESHOLD = WINDOW_MAX_SLOW - WINDOW - 2
|
|
|
|
# If the RTT rate is higher than this value,
|
|
# the max window size for fast links will be used.
|
|
# The default is 50 Kbps (the value is stored in
|
|
# bytes per second, hence the "/ 8").
|
|
RATE_FAST = (50*1000) / 8
|
|
|
|
# The minimum allowed flexibility of the window size.
|
|
# The difference between window_max and window_min
|
|
# will never be smaller than this value.
|
|
WINDOW_FLEXIBILITY = 4
|
|
|
|
# Number of bytes in a map hash
|
|
MAPHASH_LEN = 4
|
|
SDU = RNS.Packet.MDU
|
|
RANDOM_HASH_SIZE = 4
|
|
|
|
# This is an indication of what the
|
|
# maximum size a resource should be, if
|
|
# it is to be handled within reasonable
|
|
# time constraint, even on small systems.
|
|
#
|
|
# A small system in this regard is
|
|
# defined as a Raspberry Pi, which should
|
|
# be able to compress, encrypt and hash-map
|
|
# the resource in about 10 seconds.
|
|
#
|
|
# This constant will be used when determining
|
|
# how to sequence the sending of large resources.
|
|
#
|
|
# Capped at 16777215 (0xFFFFFF) per segment to
|
|
# fit in 3 bytes in resource advertisements.
|
|
MAX_EFFICIENT_SIZE = 16 * 1024 * 1024 - 1
|
|
RESPONSE_MAX_GRACE_TIME = 10
|
|
|
|
# The maximum size to auto-compress with
|
|
# bz2 before sending.
|
|
AUTO_COMPRESS_MAX_SIZE = MAX_EFFICIENT_SIZE
|
|
|
|
PART_TIMEOUT_FACTOR = 4
|
|
PART_TIMEOUT_FACTOR_AFTER_RTT = 2
|
|
MAX_RETRIES = 8
|
|
MAX_ADV_RETRIES = 4
|
|
SENDER_GRACE_TIME = 10
|
|
RETRY_GRACE_TIME = 0.25
|
|
PER_RETRY_DELAY = 0.5
|
|
|
|
WATCHDOG_MAX_SLEEP = 1
|
|
|
|
HASHMAP_IS_NOT_EXHAUSTED = 0x00
|
|
HASHMAP_IS_EXHAUSTED = 0xFF
|
|
|
|
# Status constants
|
|
NONE = 0x00
|
|
QUEUED = 0x01
|
|
ADVERTISED = 0x02
|
|
TRANSFERRING = 0x03
|
|
AWAITING_PROOF = 0x04
|
|
ASSEMBLING = 0x05
|
|
COMPLETE = 0x06
|
|
FAILED = 0x07
|
|
CORRUPT = 0x08
|
|
|
|
@staticmethod
|
|
def accept(advertisement_packet, callback=None, progress_callback = None, request_id = None):
|
|
try:
|
|
adv = ResourceAdvertisement.unpack(advertisement_packet.plaintext)
|
|
|
|
resource = Resource(None, advertisement_packet.link, request_id = request_id)
|
|
resource.status = Resource.TRANSFERRING
|
|
|
|
resource.flags = adv.f
|
|
resource.size = adv.t
|
|
resource.total_size = adv.d
|
|
resource.uncompressed_size = adv.d
|
|
resource.hash = adv.h
|
|
resource.original_hash = adv.o
|
|
resource.random_hash = adv.r
|
|
resource.hashmap_raw = adv.m
|
|
resource.encrypted = True if resource.flags & 0x01 else False
|
|
resource.compressed = True if resource.flags >> 1 & 0x01 else False
|
|
resource.initiator = False
|
|
resource.callback = callback
|
|
resource.__progress_callback = progress_callback
|
|
resource.total_parts = int(math.ceil(resource.size/float(Resource.SDU)))
|
|
resource.received_count = 0
|
|
resource.outstanding_parts = 0
|
|
resource.parts = [None] * resource.total_parts
|
|
resource.window = Resource.WINDOW
|
|
resource.window_max = Resource.WINDOW_MAX_SLOW
|
|
resource.window_min = Resource.WINDOW_MIN
|
|
resource.window_flexibility = Resource.WINDOW_FLEXIBILITY
|
|
resource.last_activity = time.time()
|
|
|
|
resource.storagepath = RNS.Reticulum.resourcepath+"/"+resource.original_hash.hex()
|
|
resource.segment_index = adv.i
|
|
resource.total_segments = adv.l
|
|
if adv.l > 1:
|
|
resource.split = True
|
|
else:
|
|
resource.split = False
|
|
|
|
resource.hashmap = [None] * resource.total_parts
|
|
resource.hashmap_height = 0
|
|
resource.waiting_for_hmu = False
|
|
|
|
resource.receiving_part = False
|
|
|
|
resource.consecutive_completed_height = 0
|
|
|
|
if not resource.link.has_incoming_resource(resource):
|
|
resource.link.register_incoming_resource(resource)
|
|
|
|
RNS.log("Accepting resource advertisement for "+RNS.prettyhexrep(resource.hash), RNS.LOG_DEBUG)
|
|
if resource.link.callbacks.resource_started != None:
|
|
try:
|
|
resource.link.callbacks.resource_started(resource)
|
|
except Exception as e:
|
|
RNS.log("Error while executing resource started callback from "+str(resource)+". The contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
|
|
resource.hashmap_update(0, resource.hashmap_raw)
|
|
|
|
resource.watchdog_job()
|
|
|
|
return resource
|
|
|
|
else:
|
|
RNS.log("Ignoring resource advertisement for "+RNS.prettyhexrep(resource.hash)+", resource already transferring", RNS.LOG_DEBUG)
|
|
return None
|
|
|
|
except Exception as e:
|
|
RNS.log("Could not decode resource advertisement, dropping resource", RNS.LOG_DEBUG)
|
|
return None
|
|
|
|
# Create a resource for transmission to a remote destination
|
|
# The data passed can be either a bytes-array or a file opened
|
|
# in binary read mode.
|
|
def __init__(self, data, link, advertise=True, auto_compress=True, callback=None, progress_callback=None, timeout = None, segment_index = 1, original_hash = None, request_id = None, is_response = False):
|
|
data_size = None
|
|
resource_data = None
|
|
self.assembly_lock = False
|
|
|
|
if hasattr(data, "read"):
|
|
data_size = os.stat(data.name).st_size
|
|
self.total_size = data_size
|
|
self.grand_total_parts = math.ceil(data_size/Resource.SDU)
|
|
|
|
if data_size <= Resource.MAX_EFFICIENT_SIZE:
|
|
self.total_segments = 1
|
|
self.segment_index = 1
|
|
self.split = False
|
|
resource_data = data.read()
|
|
data.close()
|
|
else:
|
|
self.total_segments = ((data_size-1)//Resource.MAX_EFFICIENT_SIZE)+1
|
|
self.segment_index = segment_index
|
|
self.split = True
|
|
seek_index = segment_index-1
|
|
seek_position = seek_index*Resource.MAX_EFFICIENT_SIZE
|
|
|
|
data.seek(seek_position)
|
|
resource_data = data.read(Resource.MAX_EFFICIENT_SIZE)
|
|
self.input_file = data
|
|
|
|
elif isinstance(data, bytes):
|
|
data_size = len(data)
|
|
self.grand_total_parts = math.ceil(data_size/Resource.SDU)
|
|
self.total_size = data_size
|
|
|
|
resource_data = data
|
|
self.total_segments = 1
|
|
self.segment_index = 1
|
|
self.split = False
|
|
|
|
elif data == None:
|
|
pass
|
|
|
|
else:
|
|
raise TypeError("Invalid data instance type passed to resource initialisation")
|
|
|
|
data = resource_data
|
|
|
|
self.status = Resource.NONE
|
|
self.link = link
|
|
self.max_retries = Resource.MAX_RETRIES
|
|
self.max_adv_retries = Resource.MAX_ADV_RETRIES
|
|
self.retries_left = self.max_retries
|
|
self.timeout_factor = self.link.traffic_timeout_factor
|
|
self.part_timeout_factor = Resource.PART_TIMEOUT_FACTOR
|
|
self.sender_grace_time = Resource.SENDER_GRACE_TIME
|
|
self.hmu_retry_ok = False
|
|
self.watchdog_lock = False
|
|
self.__watchdog_job_id = 0
|
|
self.__progress_callback = progress_callback
|
|
self.rtt = None
|
|
self.rtt_rxd_bytes = 0
|
|
self.req_sent = 0
|
|
self.req_resp_rtt_rate = 0
|
|
self.rtt_rxd_bytes_at_part_req = 0
|
|
self.fast_rate_rounds = 0
|
|
self.request_id = request_id
|
|
self.is_response = is_response
|
|
|
|
self.req_hashlist = []
|
|
self.receiver_min_consecutive_height = 0
|
|
|
|
if timeout != None:
|
|
self.timeout = timeout
|
|
else:
|
|
self.timeout = self.link.rtt * self.link.traffic_timeout_factor
|
|
|
|
if data != None:
|
|
self.initiator = True
|
|
self.callback = callback
|
|
self.uncompressed_data = data
|
|
|
|
compression_began = time.time()
|
|
if (auto_compress and len(self.uncompressed_data) < Resource.AUTO_COMPRESS_MAX_SIZE):
|
|
RNS.log("Compressing resource data...", RNS.LOG_DEBUG)
|
|
self.compressed_data = bz2.compress(self.uncompressed_data)
|
|
RNS.log("Compression completed in "+str(round(time.time()-compression_began, 3))+" seconds", RNS.LOG_DEBUG)
|
|
else:
|
|
self.compressed_data = self.uncompressed_data
|
|
|
|
self.uncompressed_size = len(self.uncompressed_data)
|
|
self.compressed_size = len(self.compressed_data)
|
|
|
|
if (self.compressed_size < self.uncompressed_size and auto_compress):
|
|
saved_bytes = len(self.uncompressed_data) - len(self.compressed_data)
|
|
RNS.log("Compression saved "+str(saved_bytes)+" bytes, sending compressed", RNS.LOG_DEBUG)
|
|
|
|
self.data = b""
|
|
self.data += RNS.Identity.get_random_hash()[:Resource.RANDOM_HASH_SIZE]
|
|
self.data += self.compressed_data
|
|
|
|
self.compressed = True
|
|
self.uncompressed_data = None
|
|
|
|
else:
|
|
self.data = b""
|
|
self.data += RNS.Identity.get_random_hash()[:Resource.RANDOM_HASH_SIZE]
|
|
self.data += self.uncompressed_data
|
|
self.uncompressed_data = self.data
|
|
|
|
self.compressed = False
|
|
self.compressed_data = None
|
|
if auto_compress:
|
|
RNS.log("Compression did not decrease size, sending uncompressed", RNS.LOG_DEBUG)
|
|
|
|
# Resources handle encryption directly to
|
|
# make optimal use of packet MTU on an entire
|
|
# encrypted stream. The Resource instance will
|
|
# use it's underlying link directly to encrypt.
|
|
self.data = self.link.encrypt(self.data)
|
|
self.encrypted = True
|
|
|
|
self.size = len(self.data)
|
|
self.sent_parts = 0
|
|
hashmap_entries = int(math.ceil(self.size/float(Resource.SDU)))
|
|
|
|
hashmap_ok = False
|
|
while not hashmap_ok:
|
|
hashmap_computation_began = time.time()
|
|
RNS.log("Starting resource hashmap computation with "+str(hashmap_entries)+" entries...", RNS.LOG_DEBUG)
|
|
|
|
self.random_hash = RNS.Identity.get_random_hash()[:Resource.RANDOM_HASH_SIZE]
|
|
self.hash = RNS.Identity.full_hash(data+self.random_hash)
|
|
self.truncated_hash = RNS.Identity.truncated_hash(data+self.random_hash)
|
|
self.expected_proof = RNS.Identity.full_hash(data+self.hash)
|
|
|
|
if original_hash == None:
|
|
self.original_hash = self.hash
|
|
else:
|
|
self.original_hash = original_hash
|
|
|
|
self.parts = []
|
|
self.hashmap = b""
|
|
collision_guard_list = []
|
|
for i in range(0,hashmap_entries):
|
|
data = self.data[i*Resource.SDU:(i+1)*Resource.SDU]
|
|
map_hash = self.get_map_hash(data)
|
|
|
|
if map_hash in collision_guard_list:
|
|
RNS.log("Found hash collision in resource map, remapping...", RNS.LOG_VERBOSE)
|
|
hashmap_ok = False
|
|
break
|
|
else:
|
|
hashmap_ok = True
|
|
collision_guard_list.append(map_hash)
|
|
if len(collision_guard_list) > ResourceAdvertisement.COLLISION_GUARD_SIZE:
|
|
collision_guard_list.pop(0)
|
|
|
|
part = RNS.Packet(link, data, context=RNS.Packet.RESOURCE)
|
|
part.pack()
|
|
part.map_hash = map_hash
|
|
|
|
self.hashmap += part.map_hash
|
|
self.parts.append(part)
|
|
|
|
RNS.log("Hashmap computation concluded in "+str(round(time.time()-hashmap_computation_began, 3))+" seconds", RNS.LOG_DEBUG)
|
|
|
|
if advertise:
|
|
self.advertise()
|
|
else:
|
|
pass
|
|
|
|
def hashmap_update_packet(self, plaintext):
|
|
if not self.status == Resource.FAILED:
|
|
self.last_activity = time.time()
|
|
self.retries_left = self.max_retries
|
|
|
|
update = umsgpack.unpackb(plaintext[RNS.Identity.HASHLENGTH//8:])
|
|
self.hashmap_update(update[0], update[1])
|
|
|
|
|
|
def hashmap_update(self, segment, hashmap):
|
|
if not self.status == Resource.FAILED:
|
|
self.status = Resource.TRANSFERRING
|
|
seg_len = ResourceAdvertisement.HASHMAP_MAX_LEN
|
|
hashes = len(hashmap)//Resource.MAPHASH_LEN
|
|
for i in range(0,hashes):
|
|
if self.hashmap[i+segment*seg_len] == None:
|
|
self.hashmap_height += 1
|
|
self.hashmap[i+segment*seg_len] = hashmap[i*Resource.MAPHASH_LEN:(i+1)*Resource.MAPHASH_LEN]
|
|
|
|
self.waiting_for_hmu = False
|
|
self.request_next()
|
|
|
|
def get_map_hash(self, data):
|
|
return RNS.Identity.full_hash(data+self.random_hash)[:Resource.MAPHASH_LEN]
|
|
|
|
def advertise(self):
|
|
"""
|
|
Advertise the resource. If the other end of the link accepts
|
|
the resource advertisement it will begin transferring.
|
|
"""
|
|
thread = threading.Thread(target=self.__advertise_job)
|
|
thread.daemon = True
|
|
thread.start()
|
|
|
|
def __advertise_job(self):
|
|
self.advertisement_packet = RNS.Packet(self.link, ResourceAdvertisement(self).pack(), context=RNS.Packet.RESOURCE_ADV)
|
|
while not self.link.ready_for_new_resource():
|
|
self.status = Resource.QUEUED
|
|
sleep(0.25)
|
|
|
|
try:
|
|
self.advertisement_packet.send()
|
|
self.last_activity = time.time()
|
|
self.adv_sent = self.last_activity
|
|
self.rtt = None
|
|
self.status = Resource.ADVERTISED
|
|
self.retries_left = self.max_adv_retries
|
|
self.link.register_outgoing_resource(self)
|
|
RNS.log("Sent resource advertisement for "+RNS.prettyhexrep(self.hash), RNS.LOG_DEBUG)
|
|
except Exception as e:
|
|
RNS.log("Could not advertise resource, the contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
self.cancel()
|
|
return
|
|
|
|
self.watchdog_job()
|
|
|
|
def watchdog_job(self):
|
|
thread = threading.Thread(target=self.__watchdog_job)
|
|
thread.daemon = True
|
|
thread.start()
|
|
|
|
def __watchdog_job(self):
|
|
self.__watchdog_job_id += 1
|
|
this_job_id = self.__watchdog_job_id
|
|
|
|
while self.status < Resource.ASSEMBLING and this_job_id == self.__watchdog_job_id:
|
|
while self.watchdog_lock:
|
|
sleep(0.025)
|
|
|
|
sleep_time = None
|
|
|
|
if self.status == Resource.ADVERTISED:
|
|
sleep_time = (self.adv_sent+self.timeout)-time.time()
|
|
if sleep_time < 0:
|
|
if self.retries_left <= 0:
|
|
RNS.log("Resource transfer timeout after sending advertisement", RNS.LOG_DEBUG)
|
|
self.cancel()
|
|
sleep_time = 0.001
|
|
else:
|
|
try:
|
|
RNS.log("No part requests received, retrying resource advertisement...", RNS.LOG_DEBUG)
|
|
self.retries_left -= 1
|
|
self.advertisement_packet = RNS.Packet(self.link, ResourceAdvertisement(self).pack(), context=RNS.Packet.RESOURCE_ADV)
|
|
self.advertisement_packet.send()
|
|
self.last_activity = time.time()
|
|
self.adv_sent = self.last_activity
|
|
sleep_time = 0.001
|
|
except Exception as e:
|
|
RNS.log("Could not resend advertisement packet, cancelling resource", RNS.LOG_VERBOSE)
|
|
self.cancel()
|
|
|
|
|
|
elif self.status == Resource.TRANSFERRING:
|
|
if not self.initiator:
|
|
|
|
if self.rtt == None:
|
|
rtt = self.link.rtt
|
|
else:
|
|
rtt = self.rtt
|
|
|
|
window_remaining = self.outstanding_parts
|
|
|
|
retries_used = self.max_retries - self.retries_left
|
|
extra_wait = retries_used * Resource.PER_RETRY_DELAY
|
|
sleep_time = self.last_activity + (rtt*(self.part_timeout_factor+window_remaining)) + Resource.RETRY_GRACE_TIME + extra_wait - time.time()
|
|
|
|
if sleep_time < 0:
|
|
if self.retries_left > 0:
|
|
RNS.log("Timed out waiting for parts, requesting retry", RNS.LOG_DEBUG)
|
|
if self.window > self.window_min:
|
|
self.window -= 1
|
|
if self.window_max > self.window_min:
|
|
self.window_max -= 1
|
|
if (self.window_max - self.window) > (self.window_flexibility-1):
|
|
self.window_max -= 1
|
|
|
|
sleep_time = 0.001
|
|
self.retries_left -= 1
|
|
self.waiting_for_hmu = False
|
|
self.request_next()
|
|
else:
|
|
self.cancel()
|
|
sleep_time = 0.001
|
|
else:
|
|
max_extra_wait = sum([(r+1) * Resource.PER_RETRY_DELAY for r in range(self.MAX_RETRIES)])
|
|
max_wait = self.rtt * self.timeout_factor * self.max_retries + self.sender_grace_time + max_extra_wait
|
|
sleep_time = self.last_activity + max_wait - time.time()
|
|
if sleep_time < 0:
|
|
RNS.log("Resource timed out waiting for part requests", RNS.LOG_DEBUG)
|
|
self.cancel()
|
|
sleep_time = 0.001
|
|
|
|
elif self.status == Resource.AWAITING_PROOF:
|
|
sleep_time = self.last_part_sent + (self.rtt*self.timeout_factor+self.sender_grace_time) - time.time()
|
|
if sleep_time < 0:
|
|
if self.retries_left <= 0:
|
|
RNS.log("Resource timed out waiting for proof", RNS.LOG_DEBUG)
|
|
self.cancel()
|
|
sleep_time = 0.001
|
|
else:
|
|
RNS.log("All parts sent, but no resource proof received, querying network cache...", RNS.LOG_DEBUG)
|
|
self.retries_left -= 1
|
|
expected_data = self.hash + self.expected_proof
|
|
expected_proof_packet = RNS.Packet(self.link, expected_data, packet_type=RNS.Packet.PROOF, context=RNS.Packet.RESOURCE_PRF)
|
|
expected_proof_packet.pack()
|
|
RNS.Transport.cache_request(expected_proof_packet.packet_hash, self.link)
|
|
self.last_part_sent = time.time()
|
|
sleep_time = 0.001
|
|
|
|
if sleep_time == 0:
|
|
RNS.log("Warning! Link watchdog sleep time of 0!", RNS.LOG_WARNING)
|
|
if sleep_time == None or sleep_time < 0:
|
|
RNS.log("Timing error, cancelling resource transfer.", RNS.LOG_ERROR)
|
|
self.cancel()
|
|
|
|
if sleep_time != None:
|
|
sleep(min(sleep_time, Resource.WATCHDOG_MAX_SLEEP))
|
|
|
|
def assemble(self):
|
|
if not self.status == Resource.FAILED:
|
|
try:
|
|
self.status = Resource.ASSEMBLING
|
|
stream = b"".join(self.parts)
|
|
|
|
if self.encrypted:
|
|
data = self.link.decrypt(stream)
|
|
else:
|
|
data = stream
|
|
|
|
# Strip off random hash
|
|
data = data[Resource.RANDOM_HASH_SIZE:]
|
|
|
|
if self.compressed:
|
|
self.data = bz2.decompress(data)
|
|
else:
|
|
self.data = data
|
|
|
|
calculated_hash = RNS.Identity.full_hash(self.data+self.random_hash)
|
|
|
|
if calculated_hash == self.hash:
|
|
self.file = open(self.storagepath, "ab")
|
|
self.file.write(self.data)
|
|
self.file.close()
|
|
self.status = Resource.COMPLETE
|
|
self.prove()
|
|
else:
|
|
self.status = Resource.CORRUPT
|
|
|
|
|
|
except Exception as e:
|
|
RNS.log("Error while assembling received resource.", RNS.LOG_ERROR)
|
|
RNS.log("The contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
self.status = Resource.CORRUPT
|
|
|
|
self.link.resource_concluded(self)
|
|
|
|
if self.segment_index == self.total_segments:
|
|
if self.callback != None:
|
|
self.data = open(self.storagepath, "rb")
|
|
try:
|
|
self.callback(self)
|
|
except Exception as e:
|
|
RNS.log("Error while executing resource assembled callback from "+str(self)+". The contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
|
|
try:
|
|
if hasattr(self.data, "close") and callable(self.data.close):
|
|
self.data.close()
|
|
|
|
os.unlink(self.storagepath)
|
|
|
|
except Exception as e:
|
|
RNS.log("Error while cleaning up resource files, the contained exception was:", RNS.LOG_ERROR)
|
|
RNS.log(str(e))
|
|
else:
|
|
RNS.log("Resource segment "+str(self.segment_index)+" of "+str(self.total_segments)+" received, waiting for next segment to be announced", RNS.LOG_DEBUG)
|
|
|
|
|
|
def prove(self):
|
|
if not self.status == Resource.FAILED:
|
|
try:
|
|
proof = RNS.Identity.full_hash(self.data+self.hash)
|
|
proof_data = self.hash+proof
|
|
proof_packet = RNS.Packet(self.link, proof_data, packet_type=RNS.Packet.PROOF, context=RNS.Packet.RESOURCE_PRF)
|
|
proof_packet.send()
|
|
except Exception as e:
|
|
RNS.log("Could not send proof packet, cancelling resource", RNS.LOG_DEBUG)
|
|
RNS.log("The contained exception was: "+str(e), RNS.LOG_DEBUG)
|
|
self.cancel()
|
|
|
|
def validate_proof(self, proof_data):
|
|
if not self.status == Resource.FAILED:
|
|
if len(proof_data) == RNS.Identity.HASHLENGTH//8*2:
|
|
if proof_data[RNS.Identity.HASHLENGTH//8:] == self.expected_proof:
|
|
self.status = Resource.COMPLETE
|
|
self.link.resource_concluded(self)
|
|
if self.segment_index == self.total_segments:
|
|
# If all segments were processed, we'll
|
|
# signal that the resource sending concluded
|
|
if self.callback != None:
|
|
try:
|
|
self.callback(self)
|
|
except Exception as e:
|
|
RNS.log("Error while executing resource concluded callback from "+str(self)+". The contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
else:
|
|
# Otherwise we'll recursively create the
|
|
# next segment of the resource
|
|
Resource(self.input_file, self.link, callback = self.callback, segment_index = self.segment_index+1, original_hash=self.original_hash, progress_callback = self.__progress_callback)
|
|
else:
|
|
pass
|
|
else:
|
|
pass
|
|
|
|
|
|
def receive_part(self, packet):
|
|
while self.receiving_part:
|
|
sleep(0.001)
|
|
|
|
self.receiving_part = True
|
|
self.last_activity = time.time()
|
|
self.retries_left = self.max_retries
|
|
|
|
if self.req_resp == None:
|
|
self.req_resp = self.last_activity
|
|
rtt = self.req_resp-self.req_sent
|
|
|
|
self.part_timeout_factor = Resource.PART_TIMEOUT_FACTOR_AFTER_RTT
|
|
if self.rtt == None:
|
|
self.rtt = self.link.rtt
|
|
self.watchdog_job()
|
|
elif rtt < self.rtt:
|
|
self.rtt = max(self.rtt - self.rtt*0.05, rtt)
|
|
elif rtt > self.rtt:
|
|
self.rtt = min(self.rtt + self.rtt*0.05, rtt)
|
|
|
|
if rtt > 0:
|
|
req_resp_cost = len(packet.raw)+self.req_sent_bytes
|
|
self.req_resp_rtt_rate = req_resp_cost / rtt
|
|
|
|
if self.req_resp_rtt_rate > Resource.RATE_FAST and self.fast_rate_rounds < Resource.FAST_RATE_THRESHOLD:
|
|
self.fast_rate_rounds += 1
|
|
|
|
if self.fast_rate_rounds == Resource.FAST_RATE_THRESHOLD:
|
|
self.window_max = Resource.WINDOW_MAX_FAST
|
|
|
|
if not self.status == Resource.FAILED:
|
|
self.status = Resource.TRANSFERRING
|
|
part_data = packet.data
|
|
part_hash = self.get_map_hash(part_data)
|
|
|
|
i = self.consecutive_completed_height
|
|
for map_hash in self.hashmap[self.consecutive_completed_height:self.consecutive_completed_height+self.window]:
|
|
if map_hash == part_hash:
|
|
if self.parts[i] == None:
|
|
# Insert data into parts list
|
|
self.parts[i] = part_data
|
|
self.rtt_rxd_bytes += len(part_data)
|
|
self.received_count += 1
|
|
self.outstanding_parts -= 1
|
|
|
|
# Update consecutive completed pointer
|
|
if i == self.consecutive_completed_height + 1:
|
|
self.consecutive_completed_height = i
|
|
|
|
cp = self.consecutive_completed_height + 1
|
|
while cp < len(self.parts) and self.parts[cp] != None:
|
|
self.consecutive_completed_height = cp
|
|
cp += 1
|
|
|
|
if self.__progress_callback != None:
|
|
try:
|
|
self.__progress_callback(self)
|
|
except Exception as e:
|
|
RNS.log("Error while executing progress callback from "+str(self)+". The contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
|
|
i += 1
|
|
|
|
self.receiving_part = False
|
|
|
|
if self.received_count == self.total_parts and not self.assembly_lock:
|
|
self.assembly_lock = True
|
|
self.assemble()
|
|
elif self.outstanding_parts == 0:
|
|
# TODO: Figure out if there is a mathematically
|
|
# optimal way to adjust windows
|
|
if self.window < self.window_max:
|
|
self.window += 1
|
|
if (self.window - self.window_min) > (self.window_flexibility-1):
|
|
self.window_min += 1
|
|
|
|
if self.req_sent != 0:
|
|
rtt = time.time()-self.req_sent
|
|
req_transferred = self.rtt_rxd_bytes - self.rtt_rxd_bytes_at_part_req
|
|
|
|
if rtt != 0:
|
|
self.req_data_rtt_rate = req_transferred/rtt
|
|
self.rtt_rxd_bytes_at_part_req = self.rtt_rxd_bytes
|
|
|
|
if self.req_data_rtt_rate > Resource.RATE_FAST and self.fast_rate_rounds < Resource.FAST_RATE_THRESHOLD:
|
|
self.fast_rate_rounds += 1
|
|
|
|
if self.fast_rate_rounds == Resource.FAST_RATE_THRESHOLD:
|
|
self.window_max = Resource.WINDOW_MAX_FAST
|
|
|
|
self.request_next()
|
|
else:
|
|
self.receiving_part = False
|
|
|
|
# Called on incoming resource to send a request for more data
|
|
def request_next(self):
|
|
while self.receiving_part:
|
|
sleep(0.001)
|
|
|
|
if not self.status == Resource.FAILED:
|
|
if not self.waiting_for_hmu:
|
|
self.outstanding_parts = 0
|
|
hashmap_exhausted = Resource.HASHMAP_IS_NOT_EXHAUSTED
|
|
requested_hashes = b""
|
|
|
|
offset = (1 if self.consecutive_completed_height > 0 else 0)
|
|
i = 0; pn = self.consecutive_completed_height+offset
|
|
search_start = pn
|
|
|
|
for part in self.parts[search_start:search_start+self.window]:
|
|
if part == None:
|
|
part_hash = self.hashmap[pn]
|
|
if part_hash != None:
|
|
requested_hashes += part_hash
|
|
self.outstanding_parts += 1
|
|
i += 1
|
|
else:
|
|
hashmap_exhausted = Resource.HASHMAP_IS_EXHAUSTED
|
|
|
|
pn += 1
|
|
if i >= self.window or hashmap_exhausted == Resource.HASHMAP_IS_EXHAUSTED:
|
|
break
|
|
|
|
hmu_part = bytes([hashmap_exhausted])
|
|
if hashmap_exhausted == Resource.HASHMAP_IS_EXHAUSTED:
|
|
last_map_hash = self.hashmap[self.hashmap_height-1]
|
|
hmu_part += last_map_hash
|
|
self.waiting_for_hmu = True
|
|
|
|
requested_data = b""
|
|
request_data = hmu_part + self.hash + requested_hashes
|
|
request_packet = RNS.Packet(self.link, request_data, context = RNS.Packet.RESOURCE_REQ)
|
|
|
|
try:
|
|
request_packet.send()
|
|
self.last_activity = time.time()
|
|
self.req_sent = self.last_activity
|
|
self.req_sent_bytes = len(request_packet.raw)
|
|
self.req_resp = None
|
|
except Exception as e:
|
|
RNS.log("Could not send resource request packet, cancelling resource", RNS.LOG_DEBUG)
|
|
RNS.log("The contained exception was: "+str(e), RNS.LOG_DEBUG)
|
|
self.cancel()
|
|
|
|
# Called on outgoing resource to make it send more data
|
|
def request(self, request_data):
|
|
if not self.status == Resource.FAILED:
|
|
rtt = time.time() - self.adv_sent
|
|
if self.rtt == None:
|
|
self.rtt = rtt
|
|
|
|
if self.status != Resource.TRANSFERRING:
|
|
self.status = Resource.TRANSFERRING
|
|
self.watchdog_job()
|
|
|
|
self.retries_left = self.max_retries
|
|
|
|
wants_more_hashmap = True if request_data[0] == Resource.HASHMAP_IS_EXHAUSTED else False
|
|
pad = 1+Resource.MAPHASH_LEN if wants_more_hashmap else 1
|
|
|
|
requested_hashes = request_data[pad+RNS.Identity.HASHLENGTH//8:]
|
|
|
|
# Define the search scope
|
|
search_start = self.receiver_min_consecutive_height
|
|
search_end = self.receiver_min_consecutive_height+ResourceAdvertisement.COLLISION_GUARD_SIZE
|
|
|
|
map_hashes = []
|
|
for i in range(0,len(requested_hashes)//Resource.MAPHASH_LEN):
|
|
map_hash = requested_hashes[i*Resource.MAPHASH_LEN:(i+1)*Resource.MAPHASH_LEN]
|
|
map_hashes.append(map_hash)
|
|
|
|
search_scope = self.parts[search_start:search_end]
|
|
requested_parts = list(filter(lambda part: part.map_hash in map_hashes, search_scope))
|
|
|
|
for part in requested_parts:
|
|
try:
|
|
if not part.sent:
|
|
part.send()
|
|
self.sent_parts += 1
|
|
else:
|
|
part.resend()
|
|
|
|
self.last_activity = time.time()
|
|
self.last_part_sent = self.last_activity
|
|
|
|
except Exception as e:
|
|
RNS.log("Resource could not send parts, cancelling transfer!", RNS.LOG_DEBUG)
|
|
RNS.log("The contained exception was: "+str(e), RNS.LOG_DEBUG)
|
|
self.cancel()
|
|
|
|
if wants_more_hashmap:
|
|
last_map_hash = request_data[1:Resource.MAPHASH_LEN+1]
|
|
|
|
part_index = self.receiver_min_consecutive_height
|
|
search_start = part_index
|
|
search_end = self.receiver_min_consecutive_height+ResourceAdvertisement.COLLISION_GUARD_SIZE
|
|
for part in self.parts[search_start:search_end]:
|
|
part_index += 1
|
|
if part.map_hash == last_map_hash:
|
|
break
|
|
|
|
self.receiver_min_consecutive_height = max(part_index-1-Resource.WINDOW_MAX, 0)
|
|
|
|
if part_index % ResourceAdvertisement.HASHMAP_MAX_LEN != 0:
|
|
RNS.log("Resource sequencing error, cancelling transfer!", RNS.LOG_ERROR)
|
|
self.cancel()
|
|
return
|
|
else:
|
|
segment = part_index // ResourceAdvertisement.HASHMAP_MAX_LEN
|
|
|
|
|
|
hashmap_start = segment*ResourceAdvertisement.HASHMAP_MAX_LEN
|
|
hashmap_end = min((segment+1)*ResourceAdvertisement.HASHMAP_MAX_LEN, len(self.parts))
|
|
|
|
hashmap = b""
|
|
for i in range(hashmap_start,hashmap_end):
|
|
hashmap += self.hashmap[i*Resource.MAPHASH_LEN:(i+1)*Resource.MAPHASH_LEN]
|
|
|
|
hmu = self.hash+umsgpack.packb([segment, hashmap])
|
|
hmu_packet = RNS.Packet(self.link, hmu, context = RNS.Packet.RESOURCE_HMU)
|
|
|
|
try:
|
|
hmu_packet.send()
|
|
self.last_activity = time.time()
|
|
except Exception as e:
|
|
RNS.log("Could not send resource HMU packet, cancelling resource", RNS.LOG_DEBUG)
|
|
RNS.log("The contained exception was: "+str(e), RNS.LOG_DEBUG)
|
|
self.cancel()
|
|
|
|
if self.sent_parts == len(self.parts):
|
|
self.status = Resource.AWAITING_PROOF
|
|
|
|
if self.__progress_callback != None:
|
|
try:
|
|
self.__progress_callback(self)
|
|
except Exception as e:
|
|
RNS.log("Error while executing progress callback from "+str(self)+". The contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
|
|
def cancel(self):
|
|
"""
|
|
Cancels transferring the resource.
|
|
"""
|
|
if self.status < Resource.COMPLETE:
|
|
self.status = Resource.FAILED
|
|
if self.initiator:
|
|
if self.link.status == RNS.Link.ACTIVE:
|
|
try:
|
|
cancel_packet = RNS.Packet(self.link, self.hash, context=RNS.Packet.RESOURCE_ICL)
|
|
cancel_packet.send()
|
|
except Exception as e:
|
|
RNS.log("Could not send resource cancel packet, the contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
self.link.cancel_outgoing_resource(self)
|
|
else:
|
|
self.link.cancel_incoming_resource(self)
|
|
|
|
if self.callback != None:
|
|
try:
|
|
self.link.resource_concluded(self)
|
|
self.callback(self)
|
|
except Exception as e:
|
|
RNS.log("Error while executing callbacks on resource cancel from "+str(self)+". The contained exception was: "+str(e), RNS.LOG_ERROR)
|
|
|
|
def set_callback(self, callback):
|
|
self.callback = callback
|
|
|
|
def progress_callback(self, callback):
|
|
self.__progress_callback = callback
|
|
|
|
def get_progress(self):
|
|
"""
|
|
:returns: The current progress of the resource transfer as a *float* between 0.0 and 1.0.
|
|
"""
|
|
if self.initiator:
|
|
self.processed_parts = (self.segment_index-1)*math.ceil(Resource.MAX_EFFICIENT_SIZE/Resource.SDU)
|
|
self.processed_parts += self.sent_parts
|
|
self.progress_total_parts = float(self.grand_total_parts)
|
|
else:
|
|
self.processed_parts = (self.segment_index-1)*math.ceil(Resource.MAX_EFFICIENT_SIZE/Resource.SDU)
|
|
self.processed_parts += self.received_count
|
|
if self.split:
|
|
self.progress_total_parts = float(math.ceil(self.total_size/Resource.SDU))
|
|
else:
|
|
self.progress_total_parts = float(self.total_parts)
|
|
|
|
|
|
progress = self.processed_parts / self.progress_total_parts
|
|
return progress
|
|
|
|
def get_transfer_size(self):
|
|
"""
|
|
:returns: The number of bytes needed to transfer the resource.
|
|
"""
|
|
return self.size
|
|
|
|
def get_data_size(self):
|
|
"""
|
|
:returns: The total data size of the resource.
|
|
"""
|
|
return self.total_size
|
|
|
|
def get_parts(self):
|
|
"""
|
|
:returns: The number of parts the resource will be transferred in.
|
|
"""
|
|
return self.total_parts
|
|
|
|
def get_segments(self):
|
|
"""
|
|
:returns: The number of segments the resource is divided into.
|
|
"""
|
|
return self.total_segments
|
|
|
|
def get_hash(self):
|
|
"""
|
|
:returns: The hash of the resource.
|
|
"""
|
|
return self.hash
|
|
|
|
def is_compressed(self):
|
|
"""
|
|
:returns: Whether the resource is compressed.
|
|
"""
|
|
return self.compressed
|
|
|
|
def __str__(self):
|
|
return "<"+RNS.hexrep(self.hash,delimit=False)+"/"+RNS.hexrep(self.link.link_id,delimit=False)+">"
|
|
|
|
|
|
class ResourceAdvertisement:
|
|
OVERHEAD = 134
|
|
HASHMAP_MAX_LEN = math.floor((RNS.Link.MDU-OVERHEAD)/Resource.MAPHASH_LEN)
|
|
COLLISION_GUARD_SIZE = 2*Resource.WINDOW_MAX+HASHMAP_MAX_LEN
|
|
|
|
assert HASHMAP_MAX_LEN > 0, "The configured MTU is too small to include any map hashes in resource advertisments"
|
|
|
|
@staticmethod
|
|
def is_request(advertisement_packet):
|
|
adv = ResourceAdvertisement.unpack(advertisement_packet.plaintext)
|
|
if adv.q != None and adv.u:
|
|
return True
|
|
else:
|
|
return False
|
|
|
|
|
|
@staticmethod
|
|
def is_response(advertisement_packet):
|
|
adv = ResourceAdvertisement.unpack(advertisement_packet.plaintext)
|
|
|
|
if adv.q != None and adv.p:
|
|
return True
|
|
else:
|
|
return False
|
|
|
|
|
|
@staticmethod
|
|
def read_request_id(advertisement_packet):
|
|
adv = ResourceAdvertisement.unpack(advertisement_packet.plaintext)
|
|
return adv.q
|
|
|
|
|
|
@staticmethod
|
|
def read_transfer_size(advertisement_packet):
|
|
adv = ResourceAdvertisement.unpack(advertisement_packet.plaintext)
|
|
return adv.t
|
|
|
|
|
|
@staticmethod
|
|
def read_size(advertisement_packet):
|
|
adv = ResourceAdvertisement.unpack(advertisement_packet.plaintext)
|
|
return adv.d
|
|
|
|
|
|
def __init__(self, resource=None, request_id=None, is_response=False):
|
|
if resource != None:
|
|
self.t = resource.size # Transfer size
|
|
self.d = resource.total_size # Total uncompressed data size
|
|
self.n = len(resource.parts) # Number of parts
|
|
self.h = resource.hash # Resource hash
|
|
self.r = resource.random_hash # Resource random hash
|
|
self.o = resource.original_hash # First-segment hash
|
|
self.m = resource.hashmap # Resource hashmap
|
|
self.c = resource.compressed # Compression flag
|
|
self.e = resource.encrypted # Encryption flag
|
|
self.s = resource.split # Split flag
|
|
self.i = resource.segment_index # Segment index
|
|
self.l = resource.total_segments # Total segments
|
|
self.q = resource.request_id # ID of associated request
|
|
self.u = False # Is request flag
|
|
self.p = False # Is response flag
|
|
|
|
if self.q != None:
|
|
if not resource.is_response:
|
|
self.u = True
|
|
self.p = False
|
|
else:
|
|
self.u = False
|
|
self.p = True
|
|
|
|
# Flags
|
|
self.f = 0x00 | self.p << 4 | self.u << 3 | self.s << 2 | self.c << 1 | self.e
|
|
|
|
def get_transfer_size(self):
|
|
return self.t
|
|
|
|
def get_data_size(self):
|
|
return self.d
|
|
|
|
def get_parts(self):
|
|
return self.n
|
|
|
|
def get_segments(self):
|
|
return self.l
|
|
|
|
def get_hash(self):
|
|
return self.h
|
|
|
|
def is_compressed(self):
|
|
return self.c
|
|
|
|
def pack(self, segment=0):
|
|
hashmap_start = segment*ResourceAdvertisement.HASHMAP_MAX_LEN
|
|
hashmap_end = min((segment+1)*(ResourceAdvertisement.HASHMAP_MAX_LEN), self.n)
|
|
|
|
hashmap = b""
|
|
for i in range(hashmap_start,hashmap_end):
|
|
hashmap += self.m[i*Resource.MAPHASH_LEN:(i+1)*Resource.MAPHASH_LEN]
|
|
|
|
dictionary = {
|
|
"t": self.t, # Transfer size
|
|
"d": self.d, # Data size
|
|
"n": self.n, # Number of parts
|
|
"h": self.h, # Resource hash
|
|
"r": self.r, # Resource random hash
|
|
"o": self.o, # Original hash
|
|
"i": self.i, # Segment index
|
|
"l": self.l, # Total segments
|
|
"q": self.q, # Request ID
|
|
"f": self.f, # Resource flags
|
|
"m": hashmap
|
|
}
|
|
|
|
return umsgpack.packb(dictionary)
|
|
|
|
|
|
@staticmethod
|
|
def unpack(data):
|
|
dictionary = umsgpack.unpackb(data)
|
|
|
|
adv = ResourceAdvertisement()
|
|
adv.t = dictionary["t"]
|
|
adv.d = dictionary["d"]
|
|
adv.n = dictionary["n"]
|
|
adv.h = dictionary["h"]
|
|
adv.r = dictionary["r"]
|
|
adv.o = dictionary["o"]
|
|
adv.m = dictionary["m"]
|
|
adv.f = dictionary["f"]
|
|
adv.i = dictionary["i"]
|
|
adv.l = dictionary["l"]
|
|
adv.q = dictionary["q"]
|
|
adv.e = True if (adv.f & 0x01) == 0x01 else False
|
|
adv.c = True if ((adv.f >> 1) & 0x01) == 0x01 else False
|
|
adv.s = True if ((adv.f >> 2) & 0x01) == 0x01 else False
|
|
adv.u = True if ((adv.f >> 3) & 0x01) == 0x01 else False
|
|
adv.p = True if ((adv.f >> 4) & 0x01) == 0x01 else False
|
|
|
|
return adv |