25
from bisect import bisect
26
26
from cStringIO import StringIO
29
from bzrlib import errors
30
from bzrlib.trace import mutter
33
class ResponseRange(object):
34
"""A range in a RangeFile-object."""
36
__slots__ = ['_ent_start', '_ent_end', '_data_start']
38
def __init__(self, ent_start, ent_end, data_start):
39
self._ent_start = ent_start
40
self._ent_end = ent_end
41
self._data_start = data_start
43
def __cmp__(self, other):
44
"""Compare this to other.
46
We need this both for sorting, and so that we can
47
bisect the list of ranges.
49
if isinstance(other, int):
50
# Later on we bisect for a starting point
51
# so we allow comparing against a single integer
52
return cmp(self._ent_start, other)
54
return cmp((self._ent_start, self._ent_end, self._data_start),
55
(other._ent_start, other._ent_end, other._data_start))
58
return "%s(%s-%s,%s)" % (self.__class__.__name__,
59
self._ent_start, self._ent_end,
36
# A RangeFile expects the following grammar (simplified to outline the
37
# assumptions we rely upon).
43
# whole_file: [content_length_header] data
45
# single_range: content_range_header data
47
# multiple_range: boundary_header boundary (content_range_header data boundary)+
65
49
class RangeFile(object):
66
50
"""File-like object that allow access to partial available data.
68
Specified by a set of ranges.
52
All accesses should happen sequentially since the acquisition occurs during
53
an http response reception (as sockets can't be seeked, we simulate the
54
seek by just reading and discarding the data).
56
The access pattern is defined by a set of ranges discovered as reading
57
progress. Only one range is available at a given time, so all accesses
58
should happen with monotonically increasing offsets.
71
def __init__(self, path, input_file):
61
# in _checked_read() below, we may have to discard several MB in the worst
62
# case. To avoid buffering that much, we read and discard by chunks
63
# instead. The underlying file is either a socket or a StringIO, so reading
64
# 8k chunks should be fine.
65
_discarded_buf_size = 8192
67
# maximum size of read requests -- used to avoid MemoryError issues in recv
68
_max_read_size = 512 * 1024
70
def __init__(self, path, infile):
73
:param path: File url, for error reports.
74
:param infile: File-like socket set at body start.
76
self._data = input_file.read()
78
def _add_range(self, ent_start, ent_end, data_start):
79
"""Add an entity range.
81
:param ent_start: Start offset of entity
82
:param ent_end: End offset of entity (inclusive)
83
:param data_start: Start offset of data in data stream.
85
self._ranges.append(ResponseRange(ent_start, ent_end, data_start))
86
self._len = max(self._len, ent_end)
88
def _finish_ranges(self):
79
# When using multi parts response, this will be set with the headers
80
# associated with the range currently read.
82
# Default to the whole file of unspecified size
85
def set_range(self, start, size):
86
"""Change the range mapping"""
89
# Set the new _pos since that's what we want to expose
90
self._pos = self._start
92
def set_boundary(self, boundary):
93
"""Define the boundary used in a multi parts message.
95
The file should be at the beginning of the body, the first range
96
definition is read and taken into account.
98
self._boundary = boundary
99
# Decode the headers and setup the first range
101
self.read_range_definition()
103
def read_boundary(self):
104
"""Read the boundary headers defining a new range"""
105
boundary_line = '\r\n'
106
while boundary_line == '\r\n':
107
# RFC2616 19.2 Additional CRLFs may precede the first boundary
109
# To be on the safe side we allow it before any boundary line
110
boundary_line = self._file.readline()
112
if boundary_line != '--' + self._boundary + '\r\n':
113
# rfc822.unquote() incorrectly unquotes strings enclosed in <>
114
# IIS 6 and 7 incorrectly wrap boundary strings in <>
115
# together they make a beautiful bug, which we will be gracious
117
if (self._unquote_boundary(boundary_line) !=
118
'--' + self._boundary + '\r\n'):
119
raise errors.InvalidHttpResponse(
121
"Expected a boundary (%s) line, got '%s'"
122
% (self._boundary, boundary_line))
124
def _unquote_boundary(self, b):
125
return b[:2] + rfc822.unquote(b[2:-2]) + b[-2:]
127
def read_range_definition(self):
128
"""Read a new range definition in a multi parts message.
130
Parse the headers including the empty line following them so that we
131
are ready to read the data itself.
133
self._headers = httplib.HTTPMessage(self._file, seekable=0)
134
# Extract the range definition
135
content_range = self._headers.getheader('content-range', None)
136
if content_range is None:
137
raise errors.InvalidHttpResponse(
139
'Content-Range header missing in a multi-part response')
140
self.set_range_from_header(content_range)
142
def set_range_from_header(self, content_range):
143
"""Helper to set the new range from its description in the headers"""
145
rtype, values = content_range.split()
147
raise errors.InvalidHttpRange(self._path, content_range,
150
raise errors.InvalidHttpRange(self._path, content_range,
151
"Unsupported range type '%s'" % rtype)
153
# We don't need total, but note that it may be either the file size
154
# or '*' if the server can't or doesn't want to return the file
156
start_end, total = values.split('/')
157
start, end = start_end.split('-')
161
raise errors.InvalidHttpRange(self._path, content_range,
162
'Invalid range values')
163
size = end - start + 1
165
raise errors.InvalidHttpRange(self._path, content_range,
166
'Invalid range, size <= 0')
167
self.set_range(start, size)
169
def _checked_read(self, size):
170
"""Read the file checking for short reads.
172
The data read is discarded along the way.
177
data = self._file.read(min(remaining, self._discarded_buf_size))
178
remaining -= len(data)
180
raise errors.ShortReadvError(self._path, pos, size,
184
def _seek_to_next_range(self):
185
# We will cross range boundaries
186
if self._boundary is None:
187
# If we don't have a boundary, we can't find another range
188
raise errors.InvalidRange(self._path, self._pos,
189
"Range (%s, %s) exhausted"
190
% (self._start, self._size))
192
self.read_range_definition()
194
def read(self, size=-1):
92
195
"""Read size bytes from the current position in the file.
94
Reading across ranges is not supported.
197
Reading across ranges is not supported. We rely on the underlying http
198
client to clean the socket if we leave bytes unread. This may occur for
199
the final boundary line of a multipart response or for any range
200
request not entirely consumed by the client (due to offset coalescing)
202
:param size: The number of bytes to read. Leave unspecified or pass
96
# find the last range which has a start <= pos
97
i = bisect(self._ranges, self._pos) - 1
99
if i < 0 or self._pos > self._ranges[i]._ent_end:
100
mutter('Bisect for pos: %s failed. Found offset: %d, ranges:%s',
101
self._pos, i, self._ranges)
102
raise errors.InvalidRange(self._path, self._pos)
106
# mutter('found range %s %s for pos %s', i, self._ranges[i], self._pos)
108
if (self._pos + size - 1) > r._ent_end:
109
raise errors.InvalidRange(self._path, self._pos)
111
start = r._data_start + (self._pos - r._ent_start)
113
# mutter("range read %d bytes at %d == %d-%d", size, self._pos,
115
self._pos += (end-start)
116
return self._data[start:end]
206
and self._pos == self._start + self._size):
210
self._seek_to_next_range()
211
elif self._pos < self._start:
212
raise errors.InvalidRange(
213
self._path, self._pos,
214
"Can't read %s bytes before range (%s, %s)"
215
% (size, self._start, self._size))
217
if size > 0 and self._pos + size > self._start + self._size:
218
raise errors.InvalidRange(
219
self._path, self._pos,
220
"Can't read %s bytes across range (%s, %s)"
221
% (size, self._start, self._size))
223
# read data from file
227
# Don't read past the range definition
228
limited = self._start + self._size - self._pos
230
limited = min(limited, size)
231
osutils.pumpfile(self._file, buffer, limited, self._max_read_size)
232
data = buffer.getvalue()
234
# Update _pos respecting the data effectively read
235
self._pos += len(data)
118
238
def seek(self, offset, whence=0):
239
start_pos = self._pos
121
242
elif whence == 1:
243
final_pos = start_pos + offset
123
244
elif whence == 2:
124
self._pos = self._len + offset
246
final_pos = self._start + self._size + offset # offset < 0
248
raise errors.InvalidRange(
249
self._path, self._pos,
250
"RangeFile: can't seek from end while size is unknown")
126
252
raise ValueError("Invalid value %s for whence." % whence)
254
if final_pos < self._pos:
255
# Can't seek backwards
256
raise errors.InvalidRange(
257
self._path, self._pos,
258
'RangeFile: trying to seek backwards to %s' % final_pos)
261
cur_limit = self._start + self._size
262
while final_pos > cur_limit:
263
# We will cross range boundaries
264
remain = cur_limit - self._pos
266
# Finish reading the current range
267
self._checked_read(remain)
268
self._seek_to_next_range()
269
cur_limit = self._start + self._size
271
size = final_pos - self._pos
272
if size > 0: # size can be < 0 if we crossed a range boundary
273
# We don't need the data, just read it and throw it away
274
self._checked_read(size)
135
class HttpRangeResponse(RangeFile):
136
"""A single-range HTTP response."""
138
# TODO: jam 20060706 Consider compiling these regexes on demand
139
_CONTENT_RANGE_RE = re.compile(
140
r'\s*([^\s]+)\s+([0-9]+)-([0-9]+)/([0-9]+)\s*$')
142
def __init__(self, path, content_range, input_file):
143
# mutter("parsing 206 non-multipart response for %s", path)
144
RangeFile.__init__(self, path, input_file)
145
start, end = self._parse_range(content_range, path)
146
self._add_range(start, end, 0)
147
self._finish_ranges()
150
def _parse_range(range, path='<unknown>'):
151
"""Parse an http Content-range header and return start + end
153
:param range: The value for Content-range
154
:param path: Provide to give better error messages.
155
:return: (start, end) A tuple of integers
157
match = HttpRangeResponse._CONTENT_RANGE_RE.match(range)
159
raise errors.InvalidHttpRange(path, range,
160
"Invalid Content-range")
162
rtype, start, end, total = match.groups()
165
raise errors.InvalidHttpRange(path, range,
166
"Unsupported range type '%s'" % (rtype,))
171
except ValueError, e:
172
raise errors.InvalidHttpRange(path, range, str(e))
177
class HttpMultipartRangeResponse(RangeFile):
178
"""A multi-range HTTP response."""
180
_CONTENT_TYPE_RE = re.compile(
181
r'^\s*multipart/byteranges\s*;\s*boundary\s*=\s*("?)([^"]*?)\1\s*$')
183
# Start with --<boundary>\r\n
184
# and ignore all headers ending in \r\n
185
# except for content-range:
186
# and find the two trailing \r\n separators
187
# indicating the start of the text
188
# TODO: jam 20060706 This requires exact conformance
189
# to the spec, we probably could relax the requirement
190
# of \r\n, and use something more like (\r?\n)
192
"^--%s(?:\r\n(?:(?:content-range:([^\r]+))|[^\r]+))+\r\n\r\n")
194
def __init__(self, path, content_type, input_file):
195
# mutter("parsing 206 multipart response for %s", path)
196
# TODO: jam 20060706 Is it valid to initialize a
197
# grandparent without initializing parent?
198
RangeFile.__init__(self, path, input_file)
200
self.boundary_regex = self._parse_boundary(content_type, path)
201
# mutter('response:\n%r', self._data)
203
for match in self.boundary_regex.finditer(self._data):
204
ent_start, ent_end = HttpRangeResponse._parse_range(match.group(1),
206
self._add_range(ent_start, ent_end, match.end())
208
self._finish_ranges()
211
def _parse_boundary(ctype, path='<unknown>'):
212
"""Parse the Content-type field.
214
This expects a multipart Content-type, and returns a
215
regex which is capable of finding the boundaries
216
in the multipart data.
218
match = HttpMultipartRangeResponse._CONTENT_TYPE_RE.match(ctype)
220
raise errors.InvalidHttpContentType(path, ctype,
221
"Expected multipart/byteranges with boundary")
223
boundary = match.group(2)
224
# mutter('multipart boundary is %s', boundary)
225
pattern = HttpMultipartRangeResponse._BOUNDARY_PATT
226
return re.compile(pattern % re.escape(boundary),
227
re.IGNORECASE | re.MULTILINE)
230
def _is_multipart(content_type):
231
return content_type.startswith('multipart/byteranges;')
234
def handle_response(url, code, headers, data):
235
"""Interpret the code & headers and return a HTTP response.
237
This is a factory method which returns an appropriate HTTP response
238
based on the code & headers it's given.
280
def handle_response(url, code, msg, data):
281
"""Interpret the code & headers and wrap the provided data in a RangeFile.
283
This is a factory method which returns an appropriate RangeFile based on
284
the code & headers it's given.
240
286
:param url: The url being processed. Mostly for error reporting
241
287
:param code: The integer HTTP response code
242
:param headers: A dict-like object that contains the HTTP response headers
288
:param msg: An HTTPMessage containing the headers for the response
243
289
:param data: A file-like object that can be read() to get the
245
:return: A file-like object that can seek()+read() the
291
:return: A file-like object that can seek()+read() the
246
292
ranges indicated by the headers.
251
content_type = headers['Content-Type']
253
# When there is no content-type header we treat
254
# the response as being of type 'application/octet-stream' as per
255
# RFC2616 section 7.2.1.
294
rfile = RangeFile(url, data)
297
size = msg.getheader('content-length', None)
302
rfile.set_range(0, size)
304
content_type = msg.getheader('content-type', None)
305
if content_type is None:
306
# When there is no content-type header we treat the response as
307
# being of type 'application/octet-stream' as per RFC2616 section
256
309
# Therefore it is obviously not multipart
257
310
content_type = 'application/octet-stream'
258
311
is_multipart = False
260
is_multipart = _is_multipart(content_type)
313
is_multipart = (msg.getmaintype() == 'multipart'
314
and msg.getsubtype() == 'byteranges')
263
317
# Full fledged multipart response
264
return HttpMultipartRangeResponse(url, content_type, data)
318
rfile.set_boundary(msg.getparam('boundary'))
266
320
# A response to a range request, but not multipart
268
content_range = headers['Content-Range']
321
content_range = msg.getheader('content-range', None)
322
if content_range is None:
270
323
raise errors.InvalidHttpResponse(url,
271
324
'Missing the Content-Range header in a 206 range response')
272
return HttpRangeResponse(url, content_range, data)
274
# A regular non-range response, unfortunately the result from
275
# urllib doesn't support seek, so we wrap it in a StringIO
276
tell = getattr(data, 'tell', None)
278
return StringIO(data.read())
281
raise errors.NoSuchFile(url)
283
# We don't know which, but one of the ranges we specified
284
# was wrong. So we raise with 0 for a lack of a better
286
raise errors.InvalidRange(url,0)
288
# TODO: jam 20060713 Properly handle redirects (302 Found, etc)
289
# The '_get' code says to follow redirects, we probably
290
# should actually handle the return values
325
rfile.set_range_from_header(content_range)
292
raise errors.InvalidHttpResponse(url, "Unknown response code %s"
327
raise errors.InvalidHttpResponse(url,
328
'Unknown response code %s' % code)