1
# Copyright (C) 2006 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""RemoteTransport client for the smart-server.
19
This module shouldn't be accessed directly. The classes defined here should be
20
imported from bzrlib.smart.
23
__all__ = ['RemoteTransport', 'RemoteTCPTransport', 'RemoteSSHTransport']
25
from cStringIO import StringIO
36
from bzrlib.smart import client, medium
37
from bzrlib.symbol_versioning import (deprecated_method, one_four)
40
class _SmartStat(object):
42
def __init__(self, size, mode):
47
class RemoteTransport(transport.ConnectedTransport):
48
"""Connection to a smart server.
50
The connection holds references to the medium that can be used to send
51
requests to the server.
53
The connection has a notion of the current directory to which it's
54
connected; this is incorporated in filenames passed to the server.
56
This supports some higher-level RPC operations and can also be treated
57
like a Transport to do file-like operations.
59
The connection can be made over a tcp socket, an ssh pipe or a series of
60
http requests. There are concrete subclasses for each type:
61
RemoteTCPTransport, etc.
64
# When making a readv request, cap it at requesting 5MB of data
65
_max_readv_bytes = 5*1024*1024
67
# IMPORTANT FOR IMPLEMENTORS: RemoteTransport MUST NOT be given encoding
68
# responsibilities: Put those on SmartClient or similar. This is vital for
69
# the ability to support multiple versions of the smart protocol over time:
70
# RemoteTransport is an adapter from the Transport object model to the
71
# SmartClient model, not an encoder.
73
# FIXME: the medium parameter should be private, only the tests requires
74
# it. It may be even clearer to define a TestRemoteTransport that handles
75
# the specific cases of providing a _client and/or a _medium, and leave
76
# RemoteTransport as an abstract class.
77
def __init__(self, url, _from_transport=None, medium=None, _client=None):
80
:param _from_transport: Another RemoteTransport instance that this
81
one is being cloned from. Attributes such as the medium will
84
:param medium: The medium to use for this RemoteTransport. If None,
85
the medium from the _from_transport is shared. If both this
86
and _from_transport are None, a new medium will be built.
87
_from_transport and medium cannot both be specified.
89
:param _client: Override the _SmartClient used by this transport. This
90
should only be used for testing purposes; normally this is
91
determined from the medium.
93
super(RemoteTransport, self).__init__(url,
94
_from_transport=_from_transport)
96
# The medium is the connection, except when we need to share it with
97
# other objects (RemoteBzrDir, RemoteRepository etc). In these cases
98
# what we want to share is really the shared connection.
100
if _from_transport is None:
101
# If no _from_transport is specified, we need to intialize the
105
medium, credentials = self._build_medium()
106
if 'hpss' in debug.debug_flags:
107
trace.mutter('hpss: Built a new medium: %s',
108
medium.__class__.__name__)
109
self._shared_connection = transport._SharedConnection(medium,
113
# No medium was specified, so share the medium from the
115
medium = self._shared_connection.connection
117
raise AssertionError(
118
"Both _from_transport (%r) and medium (%r) passed to "
119
"RemoteTransport.__init__, but these parameters are mutally "
120
"exclusive." % (_from_transport, medium))
123
self._client = client._SmartClient(medium)
125
self._client = _client
127
def _build_medium(self):
128
"""Create the medium if _from_transport does not provide one.
130
The medium is analogous to the connection for ConnectedTransport: it
131
allows connection sharing.
136
def is_readonly(self):
137
"""Smart server transport can do read/write file operations."""
139
resp = self._call2('Transport.is_readonly')
140
except errors.UnknownSmartMethod:
141
# XXX: nasty hack: servers before 0.16 don't have a
142
# 'Transport.is_readonly' verb, so we do what clients before 0.16
145
if resp == ('yes', ):
147
elif resp == ('no', ):
150
raise errors.UnexpectedSmartServerResponse(resp)
152
def get_smart_client(self):
153
return self._get_connection()
155
def get_smart_medium(self):
156
return self._get_connection()
158
@deprecated_method(one_four)
159
def get_shared_medium(self):
160
return self._get_shared_connection()
162
def _remote_path(self, relpath):
163
"""Returns the Unicode version of the absolute path for relpath."""
164
return self._combine_paths(self._path, relpath)
166
def _call(self, method, *args):
167
resp = self._call2(method, *args)
168
self._ensure_ok(resp)
170
def _call2(self, method, *args):
171
"""Call a method on the remote server."""
173
return self._client.call(method, *args)
174
except errors.ErrorFromSmartServer, err:
175
# The first argument, if present, is always a path.
177
context = {'orig_path': args[0]}
180
self._translate_error(err, **context)
182
def _call_with_body_bytes(self, method, args, body):
183
"""Call a method on the remote server with body bytes."""
185
return self._client.call_with_body_bytes(method, args, body)
186
except errors.ErrorFromSmartServer, err:
187
self._translate_error(err)
189
def has(self, relpath):
190
"""Indicate whether a remote file of the given name exists or not.
192
:see: Transport.has()
194
resp = self._call2('has', self._remote_path(relpath))
195
if resp == ('yes', ):
197
elif resp == ('no', ):
200
raise errors.UnexpectedSmartServerResponse(resp)
202
def get(self, relpath):
203
"""Return file-like object reading the contents of a remote file.
205
:see: Transport.get_bytes()/get_file()
207
return StringIO(self.get_bytes(relpath))
209
def get_bytes(self, relpath):
210
remote = self._remote_path(relpath)
212
resp, response_handler = self._client.call_expecting_body('get', remote)
213
except errors.ErrorFromSmartServer, err:
214
self._translate_error(err, relpath)
216
response_handler.cancel_read_body()
217
raise errors.UnexpectedSmartServerResponse(resp)
218
return response_handler.read_body_bytes()
220
def _serialise_optional_mode(self, mode):
226
def mkdir(self, relpath, mode=None):
227
resp = self._call2('mkdir', self._remote_path(relpath),
228
self._serialise_optional_mode(mode))
230
def open_write_stream(self, relpath, mode=None):
231
"""See Transport.open_write_stream."""
232
self.put_bytes(relpath, "", mode)
233
result = transport.AppendBasedFileStream(self, relpath)
234
transport._file_streams[self.abspath(relpath)] = result
237
def put_bytes(self, relpath, upload_contents, mode=None):
238
# FIXME: upload_file is probably not safe for non-ascii characters -
239
# should probably just pass all parameters as length-delimited
241
if type(upload_contents) is unicode:
242
# Although not strictly correct, we raise UnicodeEncodeError to be
243
# compatible with other transports.
244
raise UnicodeEncodeError(
245
'undefined', upload_contents, 0, 1,
246
'put_bytes must be given bytes, not unicode.')
247
resp = self._call_with_body_bytes('put',
248
(self._remote_path(relpath), self._serialise_optional_mode(mode)),
250
self._ensure_ok(resp)
251
return len(upload_contents)
253
def put_bytes_non_atomic(self, relpath, bytes, mode=None,
254
create_parent_dir=False,
256
"""See Transport.put_bytes_non_atomic."""
257
# FIXME: no encoding in the transport!
258
create_parent_str = 'F'
259
if create_parent_dir:
260
create_parent_str = 'T'
262
resp = self._call_with_body_bytes(
264
(self._remote_path(relpath), self._serialise_optional_mode(mode),
265
create_parent_str, self._serialise_optional_mode(dir_mode)),
267
self._ensure_ok(resp)
269
def put_file(self, relpath, upload_file, mode=None):
270
# its not ideal to seek back, but currently put_non_atomic_file depends
271
# on transports not reading before failing - which is a faulty
272
# assumption I think - RBC 20060915
273
pos = upload_file.tell()
275
return self.put_bytes(relpath, upload_file.read(), mode)
277
upload_file.seek(pos)
280
def put_file_non_atomic(self, relpath, f, mode=None,
281
create_parent_dir=False,
283
return self.put_bytes_non_atomic(relpath, f.read(), mode=mode,
284
create_parent_dir=create_parent_dir,
287
def append_file(self, relpath, from_file, mode=None):
288
return self.append_bytes(relpath, from_file.read(), mode)
290
def append_bytes(self, relpath, bytes, mode=None):
291
resp = self._call_with_body_bytes(
293
(self._remote_path(relpath), self._serialise_optional_mode(mode)),
295
if resp[0] == 'appended':
297
raise errors.UnexpectedSmartServerResponse(resp)
299
def delete(self, relpath):
300
resp = self._call2('delete', self._remote_path(relpath))
301
self._ensure_ok(resp)
303
def external_url(self):
304
"""See bzrlib.transport.Transport.external_url."""
305
# the external path for RemoteTransports is the base
308
def recommended_page_size(self):
309
"""Return the recommended page size for this transport."""
312
def _readv(self, relpath, offsets):
316
offsets = list(offsets)
318
sorted_offsets = sorted(offsets)
319
coalesced = list(self._coalesce_offsets(sorted_offsets,
320
limit=self._max_readv_combine,
321
fudge_factor=self._bytes_to_read_before_seek))
323
# now that we've coallesced things, avoid making enormous requests
328
if c.length + cur_len > self._max_readv_bytes:
329
requests.append(cur_request)
333
cur_request.append(c)
336
requests.append(cur_request)
337
if 'hpss' in debug.debug_flags:
338
trace.mutter('%s.readv %s offsets => %s coalesced'
339
' => %s requests (%s)',
340
self.__class__.__name__, len(offsets), len(coalesced),
341
len(requests), sum(map(len, requests)))
342
# Cache the results, but only until they have been fulfilled
344
# turn the list of offsets into a single stack to iterate
345
offset_stack = iter(offsets)
346
# using a list so it can be modified when passing down and coming back
347
next_offset = [offset_stack.next()]
348
for cur_request in requests:
350
result = self._client.call_with_body_readv_array(
351
('readv', self._remote_path(relpath),),
352
[(c.start, c.length) for c in cur_request])
353
resp, response_handler = result
354
except errors.ErrorFromSmartServer, err:
355
self._translate_error(err)
357
if resp[0] != 'readv':
358
# This should raise an exception
359
response_handler.cancel_read_body()
360
raise errors.UnexpectedSmartServerResponse(resp)
362
for res in self._handle_response(offset_stack, cur_request,
368
def _handle_response(self, offset_stack, coalesced, response_handler,
369
data_map, next_offset):
370
cur_offset_and_size = next_offset[0]
371
# FIXME: this should know how many bytes are needed, for clarity.
372
data = response_handler.read_body_bytes()
374
for c_offset in coalesced:
375
if len(data) < c_offset.length:
376
raise errors.ShortReadvError(relpath, c_offset.start,
377
c_offset.length, actual=len(data))
378
for suboffset, subsize in c_offset.ranges:
379
key = (c_offset.start+suboffset, subsize)
380
this_data = data[data_offset+suboffset:
381
data_offset+suboffset+subsize]
382
# Special case when the data is in-order, rather than packing
383
# into a map and then back out again. Benchmarking shows that
384
# this has 100% hit rate, but leave in the data_map work just
386
# TODO: Could we get away with using buffer() to avoid the
387
# memory copy? Callers would need to realize they may
388
# not have a real string.
389
if key == cur_offset_and_size:
390
yield cur_offset_and_size[0], this_data
391
cur_offset_and_size = next_offset[0] = offset_stack.next()
393
data_map[key] = this_data
394
data_offset += c_offset.length
396
# Now that we've read some data, see if we can yield anything back
397
while cur_offset_and_size in data_map:
398
this_data = data_map.pop(cur_offset_and_size)
399
yield cur_offset_and_size[0], this_data
400
cur_offset_and_size = next_offset[0] = offset_stack.next()
402
def rename(self, rel_from, rel_to):
404
self._remote_path(rel_from),
405
self._remote_path(rel_to))
407
def move(self, rel_from, rel_to):
409
self._remote_path(rel_from),
410
self._remote_path(rel_to))
412
def rmdir(self, relpath):
413
resp = self._call('rmdir', self._remote_path(relpath))
415
def _ensure_ok(self, resp):
417
raise errors.UnexpectedSmartServerResponse(resp)
419
def _translate_error(self, err, orig_path=None):
420
remote._translate_error(err, path=orig_path)
422
def disconnect(self):
423
self.get_smart_medium().disconnect()
425
def stat(self, relpath):
426
resp = self._call2('stat', self._remote_path(relpath))
427
if resp[0] == 'stat':
428
return _SmartStat(int(resp[1]), int(resp[2], 8))
429
raise errors.UnexpectedSmartServerResponse(resp)
431
## def lock_read(self, relpath):
432
## """Lock the given file for shared (read) access.
433
## :return: A lock object, which should be passed to Transport.unlock()
435
## # The old RemoteBranch ignore lock for reading, so we will
436
## # continue that tradition and return a bogus lock object.
437
## class BogusLock(object):
438
## def __init__(self, path):
442
## return BogusLock(relpath)
447
def list_dir(self, relpath):
448
resp = self._call2('list_dir', self._remote_path(relpath))
449
if resp[0] == 'names':
450
return [name.encode('ascii') for name in resp[1:]]
451
raise errors.UnexpectedSmartServerResponse(resp)
453
def iter_files_recursive(self):
454
resp = self._call2('iter_files_recursive', self._remote_path(''))
455
if resp[0] == 'names':
457
raise errors.UnexpectedSmartServerResponse(resp)
460
class RemoteTCPTransport(RemoteTransport):
461
"""Connection to smart server over plain tcp.
463
This is essentially just a factory to get 'RemoteTransport(url,
464
SmartTCPClientMedium).
467
def _build_medium(self):
468
client_medium = medium.SmartTCPClientMedium(
469
self._host, self._port, self.base)
470
return client_medium, None
473
class RemoteTCPTransportV2Only(RemoteTransport):
474
"""Connection to smart server over plain tcp with the client hard-coded to
475
assume protocol v2 and remote server version <= 1.6.
477
This should only be used for testing.
480
def _build_medium(self):
481
client_medium = medium.SmartTCPClientMedium(
482
self._host, self._port, self.base)
483
client_medium._protocol_version = 2
484
client_medium._remember_remote_is_before((1, 6))
485
return client_medium, None
488
class RemoteSSHTransport(RemoteTransport):
489
"""Connection to smart server over SSH.
491
This is essentially just a factory to get 'RemoteTransport(url,
492
SmartSSHClientMedium).
495
def _build_medium(self):
496
location_config = config.LocationConfig(self.base)
497
bzr_remote_path = location_config.get_bzr_remote_path()
500
auth = config.AuthenticationConfig()
501
user = auth.get_user('ssh', self._host, self._port)
502
client_medium = medium.SmartSSHClientMedium(self._host, self._port,
503
user, self._password, self.base,
504
bzr_remote_path=bzr_remote_path)
505
return client_medium, (user, self._password)
508
class RemoteHTTPTransport(RemoteTransport):
509
"""Just a way to connect between a bzr+http:// url and http://.
511
This connection operates slightly differently than the RemoteSSHTransport.
512
It uses a plain http:// transport underneath, which defines what remote
513
.bzr/smart URL we are connected to. From there, all paths that are sent are
514
sent as relative paths, this way, the remote side can properly
515
de-reference them, since it is likely doing rewrite rules to translate an
516
HTTP path into a local path.
519
def __init__(self, base, _from_transport=None, http_transport=None):
520
if http_transport is None:
521
# FIXME: the password may be lost here because it appears in the
522
# url only for an intial construction (when the url came from the
524
http_url = base[len('bzr+'):]
525
self._http_transport = transport.get_transport(http_url)
527
self._http_transport = http_transport
528
super(RemoteHTTPTransport, self).__init__(
529
base, _from_transport=_from_transport)
531
def _build_medium(self):
532
# We let http_transport take care of the credentials
533
return self._http_transport.get_smart_medium(), None
535
def _remote_path(self, relpath):
536
"""After connecting, HTTP Transport only deals in relative URLs."""
537
# Adjust the relpath based on which URL this smart transport is
539
http_base = urlutils.normalize_url(self.get_smart_medium().base)
540
url = urlutils.join(self.base[len('bzr+'):], relpath)
541
url = urlutils.normalize_url(url)
542
return urlutils.relative_url(http_base, url)
544
def clone(self, relative_url):
545
"""Make a new RemoteHTTPTransport related to me.
547
This is re-implemented rather than using the default
548
RemoteTransport.clone() because we must be careful about the underlying
551
Also, the cloned smart transport will POST to the same .bzr/smart
552
location as this transport (although obviously the relative paths in the
553
smart requests may be different). This is so that the server doesn't
554
have to handle .bzr/smart requests at arbitrary places inside .bzr
555
directories, just at the initial URL the user uses.
558
abs_url = self.abspath(relative_url)
561
return RemoteHTTPTransport(abs_url,
562
_from_transport=self,
563
http_transport=self._http_transport)
566
def get_test_permutations():
567
"""Return (transport, server) permutations for testing."""
568
### We may need a little more test framework support to construct an
569
### appropriate RemoteTransport in the future.
570
from bzrlib.smart import server
571
return [(RemoteTCPTransport, server.SmartTCPServer_for_testing)]