24
from __future__ import absolute_import
25
from bisect import bisect
26
27
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 == '':
113
# A timeout in the proxy server caused the response to end early.
114
# See launchpad bug 198646.
115
raise errors.HttpBoundaryMissing(
119
if boundary_line != '--' + self._boundary + '\r\n':
120
# rfc822.unquote() incorrectly unquotes strings enclosed in <>
121
# IIS 6 and 7 incorrectly wrap boundary strings in <>
122
# together they make a beautiful bug, which we will be gracious
124
if (self._unquote_boundary(boundary_line) !=
125
'--' + self._boundary + '\r\n'):
126
raise errors.InvalidHttpResponse(
128
"Expected a boundary (%s) line, got '%s'"
129
% (self._boundary, boundary_line))
131
def _unquote_boundary(self, b):
132
return b[:2] + rfc822.unquote(b[2:-2]) + b[-2:]
134
def read_range_definition(self):
135
"""Read a new range definition in a multi parts message.
137
Parse the headers including the empty line following them so that we
138
are ready to read the data itself.
140
self._headers = httplib.HTTPMessage(self._file, seekable=0)
141
# Extract the range definition
142
content_range = self._headers.getheader('content-range', None)
143
if content_range is None:
144
raise errors.InvalidHttpResponse(
146
'Content-Range header missing in a multi-part response')
147
self.set_range_from_header(content_range)
149
def set_range_from_header(self, content_range):
150
"""Helper to set the new range from its description in the headers"""
152
rtype, values = content_range.split()
154
raise errors.InvalidHttpRange(self._path, content_range,
157
raise errors.InvalidHttpRange(self._path, content_range,
158
"Unsupported range type '%s'" % rtype)
160
# We don't need total, but note that it may be either the file size
161
# or '*' if the server can't or doesn't want to return the file
163
start_end, total = values.split('/')
164
start, end = start_end.split('-')
168
raise errors.InvalidHttpRange(self._path, content_range,
169
'Invalid range values')
170
size = end - start + 1
172
raise errors.InvalidHttpRange(self._path, content_range,
173
'Invalid range, size <= 0')
174
self.set_range(start, size)
176
def _checked_read(self, size):
177
"""Read the file checking for short reads.
179
The data read is discarded along the way.
184
data = self._file.read(min(remaining, self._discarded_buf_size))
185
remaining -= len(data)
187
raise errors.ShortReadvError(self._path, pos, size,
191
def _seek_to_next_range(self):
192
# We will cross range boundaries
193
if self._boundary is None:
194
# If we don't have a boundary, we can't find another range
195
raise errors.InvalidRange(self._path, self._pos,
196
"Range (%s, %s) exhausted"
197
% (self._start, self._size))
199
self.read_range_definition()
201
def read(self, size=-1):
92
202
"""Read size bytes from the current position in the file.
94
Reading across ranges is not supported.
204
Reading across ranges is not supported. We rely on the underlying http
205
client to clean the socket if we leave bytes unread. This may occur for
206
the final boundary line of a multipart response or for any range
207
request not entirely consumed by the client (due to offset coalescing)
209
: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]
213
and self._pos == self._start + self._size):
217
self._seek_to_next_range()
218
elif self._pos < self._start:
219
raise errors.InvalidRange(
220
self._path, self._pos,
221
"Can't read %s bytes before range (%s, %s)"
222
% (size, self._start, self._size))
224
if size > 0 and self._pos + size > self._start + self._size:
225
raise errors.InvalidRange(
226
self._path, self._pos,
227
"Can't read %s bytes across range (%s, %s)"
228
% (size, self._start, self._size))
230
# read data from file
234
# Don't read past the range definition
235
limited = self._start + self._size - self._pos
237
limited = min(limited, size)
238
osutils.pumpfile(self._file, buffer, limited, self._max_read_size)
239
data = buffer.getvalue()
241
# Update _pos respecting the data effectively read
242
self._pos += len(data)
118
245
def seek(self, offset, whence=0):
246
start_pos = self._pos
121
249
elif whence == 1:
250
final_pos = start_pos + offset
123
251
elif whence == 2:
124
self._pos = self._len + offset
253
final_pos = self._start + self._size + offset # offset < 0
255
raise errors.InvalidRange(
256
self._path, self._pos,
257
"RangeFile: can't seek from end while size is unknown")
126
259
raise ValueError("Invalid value %s for whence." % whence)
261
if final_pos < self._pos:
262
# Can't seek backwards
263
raise errors.InvalidRange(
264
self._path, self._pos,
265
'RangeFile: trying to seek backwards to %s' % final_pos)
268
cur_limit = self._start + self._size
269
while final_pos > cur_limit:
270
# We will cross range boundaries
271
remain = cur_limit - self._pos
273
# Finish reading the current range
274
self._checked_read(remain)
275
self._seek_to_next_range()
276
cur_limit = self._start + self._size
278
size = final_pos - self._pos
279
if size > 0: # size can be < 0 if we crossed a range boundary
280
# We don't need the data, just read it and throw it away
281
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.
287
def handle_response(url, code, msg, data):
288
"""Interpret the code & headers and wrap the provided data in a RangeFile.
290
This is a factory method which returns an appropriate RangeFile based on
291
the code & headers it's given.
240
293
:param url: The url being processed. Mostly for error reporting
241
294
:param code: The integer HTTP response code
242
:param headers: A dict-like object that contains the HTTP response headers
295
:param msg: An HTTPMessage containing the headers for the response
243
296
:param data: A file-like object that can be read() to get the
245
:return: A file-like object that can seek()+read() the
298
:return: A file-like object that can seek()+read() the
246
299
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.
301
rfile = RangeFile(url, data)
304
size = msg.getheader('content-length', None)
309
rfile.set_range(0, size)
311
content_type = msg.getheader('content-type', None)
312
if content_type is None:
313
# When there is no content-type header we treat the response as
314
# being of type 'application/octet-stream' as per RFC2616 section
256
316
# Therefore it is obviously not multipart
257
317
content_type = 'application/octet-stream'
258
318
is_multipart = False
260
is_multipart = _is_multipart(content_type)
320
is_multipart = (msg.getmaintype() == 'multipart'
321
and msg.getsubtype() == 'byteranges')
263
324
# Full fledged multipart response
264
return HttpMultipartRangeResponse(url, content_type, data)
325
rfile.set_boundary(msg.getparam('boundary'))
266
327
# A response to a range request, but not multipart
268
content_range = headers['Content-Range']
328
content_range = msg.getheader('content-range', None)
329
if content_range is None:
270
330
raise errors.InvalidHttpResponse(url,
271
331
'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
332
rfile.set_range_from_header(content_range)
292
raise errors.InvalidHttpResponse(url, "Unknown response code %s"
334
raise errors.InvalidHttpResponse(url,
335
'Unknown response code %s' % code)
added
removed
bzrlib/transport/http/response.py
