~bzr-pqm/bzr/bzr.dev

# Copyright (C) 2005 Robey Pointer <robey@lag.net>

# Copyright (C) 2005, 2006, 2007 Canonical Ltd

#

# This program is free software; you can redistribute it and/or modify

# it under the terms of the GNU General Public License as published by

# the Free Software Foundation; either version 2 of the License, or

# (at your option) any later version.

#

# This program is distributed in the hope that it will be useful,

# but WITHOUT ANY WARRANTY; without even the implied warranty of

# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

# GNU General Public License for more details.

#

# You should have received a copy of the GNU General Public License

# along with this program; if not, write to the Free Software

# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

"""Implementation of Transport over SFTP, using paramiko."""

# TODO: Remove the transport-based lock_read and lock_write methods.  They'll

# then raise TransportNotPossible, which will break remote access to any

# formats which rely on OS-level locks.  That should be fine as those formats

# are pretty old, but these combinations may have to be removed from the test

# suite.  Those formats all date back to 0.7; so we should be able to remove

# these methods when we officially drop support for those formats.

import errno

import os

import random

import select

import socket

import stat

import sys

import time

import urllib

import urlparse

import warnings

from bzrlib import (

    errors,

    urlutils,

    )

from bzrlib.errors import (FileExists,

                           NoSuchFile, PathNotChild,

                           TransportError,

                           LockError,

                           PathError,

                           ParamikoNotPresent,

                           )

from bzrlib.osutils import pathjoin, fancy_rename, getcwd

from bzrlib.symbol_versioning import (

        deprecated_function,

        zero_ninety,

        )

from bzrlib.trace import mutter, warning

from bzrlib.transport import (

    FileFileStream,

    _file_streams,

    local,

    Server,

    ssh,

    ConnectedTransport,

    )

# Disable one particular warning that comes from paramiko in Python2.5; if

# this is emitted at the wrong time it tends to cause spurious test failures

# or at least noise in the test case::

#

# [1770/7639 in 86s, 1 known failures, 50 skipped, 2 missing features]

# test_permissions.TestSftpPermissions.test_new_files

# /var/lib/python-support/python2.5/paramiko/message.py:226: DeprecationWarning: integer argument expected, got float

#  self.packet.write(struct.pack('>I', n))

warnings.filterwarnings('ignore',

        'integer argument expected, got float',

        category=DeprecationWarning,

        module='paramiko.message')

try:

    import paramiko

except ImportError, e:

    raise ParamikoNotPresent(e)

else:

    from paramiko.sftp import (SFTP_FLAG_WRITE, SFTP_FLAG_CREATE,

                               SFTP_FLAG_EXCL, SFTP_FLAG_TRUNC,

                               CMD_HANDLE, CMD_OPEN)

    from paramiko.sftp_attr import SFTPAttributes

    from paramiko.sftp_file import SFTPFile

_paramiko_version = getattr(paramiko, '__version_info__', (0, 0, 0))

# don't use prefetch unless paramiko version >= 1.5.5 (there were bugs earlier)

_default_do_prefetch = (_paramiko_version >= (1, 5, 5))

@deprecated_function(zero_ninety)

def clear_connection_cache():

    """Remove all hosts from the SFTP connection cache.

    Primarily useful for test cases wanting to force garbage collection.

    We don't have a global connection cache anymore.

    """

class SFTPLock(object):

    """This fakes a lock in a remote location.

    A present lock is indicated just by the existence of a file.  This

    doesn't work well on all transports and they are only used in 

    deprecated storage formats.

    """

    __slots__ = ['path', 'lock_path', 'lock_file', 'transport']

    def __init__(self, path, transport):

        assert isinstance(transport, SFTPTransport)

        self.lock_file = None

        self.path = path

        self.lock_path = path + '.write-lock'

        self.transport = transport

        try:

            # RBC 20060103 FIXME should we be using private methods here ?

            abspath = transport._remote_path(self.lock_path)

            self.lock_file = transport._sftp_open_exclusive(abspath)

        except FileExists:

            raise LockError('File %r already locked' % (self.path,))

    def __del__(self):

        """Should this warn, or actually try to cleanup?"""

        if self.lock_file:

            warning("SFTPLock %r not explicitly unlocked" % (self.path,))

            self.unlock()

    def unlock(self):

        if not self.lock_file:

            return

        self.lock_file.close()

        self.lock_file = None

        try:

            self.transport.delete(self.lock_path)

        except (NoSuchFile,):

            # What specific errors should we catch here?

            pass

class SFTPTransport(ConnectedTransport):

    """Transport implementation for SFTP access."""

    _do_prefetch = _default_do_prefetch

    # TODO: jam 20060717 Conceivably these could be configurable, either

    #       by auto-tuning at run-time, or by a configuration (per host??)

    #       but the performance curve is pretty flat, so just going with

    #       reasonable defaults.

    _max_readv_combine = 200

    # Having to round trip to the server means waiting for a response,

    # so it is better to download extra bytes.

    # 8KiB had good performance for both local and remote network operations

    _bytes_to_read_before_seek = 8192

    # The sftp spec says that implementations SHOULD allow reads

    # to be at least 32K. paramiko.readv() does an async request

    # for the chunks. So we need to keep it within a single request

    # size for paramiko <= 1.6.1. paramiko 1.6.2 will probably chop

    # up the request itself, rather than us having to worry about it

    _max_request_size = 32768

    def __init__(self, base, _from_transport=None):

        assert base.startswith('sftp://')

        super(SFTPTransport, self).__init__(base,

                                            _from_transport=_from_transport)

    def _remote_path(self, relpath):

        """Return the path to be passed along the sftp protocol for relpath.

        :param relpath: is a urlencoded string.

        """

        relative = urlutils.unescape(relpath).encode('utf-8')

        remote_path = self._combine_paths(self._path, relative)

        # the initial slash should be removed from the path, and treated as a

        # homedir relative path (the path begins with a double slash if it is

        # absolute).  see draft-ietf-secsh-scp-sftp-ssh-uri-03.txt

        # RBC 20060118 we are not using this as its too user hostile. instead

        # we are following lftp and using /~/foo to mean '~/foo'

        # vila--20070602 and leave absolute paths begin with a single slash.

        if remote_path.startswith('/~/'):

            remote_path = remote_path[3:]

        elif remote_path == '/~':

            remote_path = ''

        return remote_path

    def _create_connection(self, credentials=None):

        """Create a new connection with the provided credentials.

        :param credentials: The credentials needed to establish the connection.

        :return: The created connection and its associated credentials.

        The credentials are only the password as it may have been entered

        interactively by the user and may be different from the one provided

        in base url at transport creation time.

        """

        if credentials is None:

            password = self._password

        else:

            password = credentials

        vendor = ssh._get_ssh_vendor()

        connection = vendor.connect_sftp(self._user, password,

                                         self._host, self._port)

        return connection, password

    def _get_sftp(self):

        """Ensures that a connection is established"""

        connection = self._get_connection()

        if connection is None:

            # First connection ever

            connection, credentials = self._create_connection()

            self._set_connection(connection, credentials)

        return connection

    def has(self, relpath):

        """

        Does the target location exist?

        """

        try:

            self._get_sftp().stat(self._remote_path(relpath))

            return True

        except IOError:

            return False

    def get(self, relpath):

        """

        Get the file at the given relative path.

        :param relpath: The relative path to the file

        """

        try:

            path = self._remote_path(relpath)

            f = self._get_sftp().file(path, mode='rb')

            if self._do_prefetch and (getattr(f, 'prefetch', None) is not None):

                f.prefetch()

            return f

        except (IOError, paramiko.SSHException), e:

            self._translate_io_exception(e, path, ': error retrieving',

                failure_exc=errors.ReadError)

    def _readv(self, relpath, offsets):

        """See Transport.readv()"""

        # We overload the default readv() because we want to use a file

        # that does not have prefetch enabled.

        # Also, if we have a new paramiko, it implements an async readv()

        if not offsets:

            return

        try:

            path = self._remote_path(relpath)

            fp = self._get_sftp().file(path, mode='rb')

            readv = getattr(fp, 'readv', None)

            if readv:

                return self._sftp_readv(fp, offsets, relpath)

            mutter('seek and read %s offsets', len(offsets))

            return self._seek_and_read(fp, offsets, relpath)

        except (IOError, paramiko.SSHException), e:

            self._translate_io_exception(e, path, ': error retrieving')

    def recommended_page_size(self):

        """See Transport.recommended_page_size().

        For SFTP we suggest a large page size to reduce the overhead

        introduced by latency.

        """

        return 64 * 1024

    def _sftp_readv(self, fp, offsets, relpath='<unknown>'):

        """Use the readv() member of fp to do async readv.

        And then read them using paramiko.readv(). paramiko.readv()

        does not support ranges > 64K, so it caps the request size, and

        just reads until it gets all the stuff it wants

        """

        offsets = list(offsets)

        sorted_offsets = sorted(offsets)

        # The algorithm works as follows:

        # 1) Coalesce nearby reads into a single chunk

        #    This generates a list of combined regions, the total size

        #    and the size of the sub regions. This coalescing step is limited

        #    in the number of nearby chunks to combine, and is allowed to

        #    skip small breaks in the requests. Limiting it makes sure that

        #    we can start yielding some data earlier, and skipping means we

        #    make fewer requests. (Beneficial even when using async)

        # 2) Break up this combined regions into chunks that are smaller

        #    than 64KiB. Technically the limit is 65536, but we are a

        #    little bit conservative. This is because sftp has a maximum

        #    return chunk size of 64KiB (max size of an unsigned short)

        # 3) Issue a readv() to paramiko to create an async request for

        #    all of this data

        # 4) Read in the data as it comes back, until we've read one

        #    continuous section as determined in step 1

        # 5) Break up the full sections into hunks for the original requested

        #    offsets. And put them in a cache

        # 6) Check if the next request is in the cache, and if it is, remove

        #    it from the cache, and yield its data. Continue until no more

        #    entries are in the cache.

        # 7) loop back to step 4 until all data has been read

        #

        # TODO: jam 20060725 This could be optimized one step further, by

        #       attempting to yield whatever data we have read, even before

        #       the first coallesced section has been fully processed.

        # When coalescing for use with readv(), we don't really need to

        # use any fudge factor, because the requests are made asynchronously

        coalesced = list(self._coalesce_offsets(sorted_offsets,

                               limit=self._max_readv_combine,

                               fudge_factor=0,

                               ))

        requests = []

        for c_offset in coalesced:

            start = c_offset.start

            size = c_offset.length

            # We need to break this up into multiple requests

            while size > 0:

                next_size = min(size, self._max_request_size)

                requests.append((start, next_size))

                size -= next_size

                start += next_size

        mutter('SFTP.readv() %s offsets => %s coalesced => %s requests',

                len(offsets), len(coalesced), len(requests))

        # Queue the current read until we have read the full coalesced section

        cur_data = []

        cur_data_len = 0

        cur_coalesced_stack = iter(coalesced)

        cur_coalesced = cur_coalesced_stack.next()

        # Cache the results, but only until they have been fulfilled

        data_map = {}

        # turn the list of offsets into a stack

        offset_stack = iter(offsets)

        cur_offset_and_size = offset_stack.next()

        for data in fp.readv(requests):

            cur_data += data

            cur_data_len += len(data)

            if cur_data_len < cur_coalesced.length:

                continue

            assert cur_data_len == cur_coalesced.length, \

                "Somehow we read too much: %s != %s" % (cur_data_len,

                                                        cur_coalesced.length)

            all_data = ''.join(cur_data)

            cur_data = []

            cur_data_len = 0

            for suboffset, subsize in cur_coalesced.ranges:

                key = (cur_coalesced.start+suboffset, subsize)

                data_map[key] = all_data[suboffset:suboffset+subsize]

            # Now that we've read some data, see if we can yield anything back

            while cur_offset_and_size in data_map:

                this_data = data_map.pop(cur_offset_and_size)

                yield cur_offset_and_size[0], this_data

                cur_offset_and_size = offset_stack.next()

            # We read a coalesced entry, so mark it as done

            cur_coalesced = None

            # Now that we've read all of the data for this coalesced section

            # on to the next

            cur_coalesced = cur_coalesced_stack.next()

        if cur_coalesced is not None:

            raise errors.ShortReadvError(relpath, cur_coalesced.start,

                cur_coalesced.length, len(data))

    def put_file(self, relpath, f, mode=None):

        """

        Copy the file-like object into the location.

        :param relpath: Location to put the contents, relative to base.

        :param f:       File-like object.

        :param mode: The final mode for the file

        """

        final_path = self._remote_path(relpath)

        return self._put(final_path, f, mode=mode)

    def _put(self, abspath, f, mode=None):

        """Helper function so both put() and copy_abspaths can reuse the code"""

        tmp_abspath = '%s.tmp.%.9f.%d.%d' % (abspath, time.time(),

                        os.getpid(), random.randint(0,0x7FFFFFFF))

        fout = self._sftp_open_exclusive(tmp_abspath, mode=mode)

        closed = False

        try:

            try:

                fout.set_pipelined(True)

                length = self._pump(f, fout)

            except (IOError, paramiko.SSHException), e:

                self._translate_io_exception(e, tmp_abspath)

            # XXX: This doesn't truly help like we would like it to.

            #      The problem is that openssh strips sticky bits. So while we

            #      can properly set group write permission, we lose the group

            #      sticky bit. So it is probably best to stop chmodding, and

            #      just tell users that they need to set the umask correctly.

            #      The attr.st_mode = mode, in _sftp_open_exclusive

            #      will handle when the user wants the final mode to be more 

            #      restrictive. And then we avoid a round trip. Unless 

            #      paramiko decides to expose an async chmod()

            # This is designed to chmod() right before we close.

            # Because we set_pipelined() earlier, theoretically we might 

            # avoid the round trip for fout.close()

            if mode is not None:

                self._get_sftp().chmod(tmp_abspath, mode)

            fout.close()

            closed = True

            self._rename_and_overwrite(tmp_abspath, abspath)

            return length

        except Exception, e:

            # If we fail, try to clean up the temporary file

            # before we throw the exception

            # but don't let another exception mess things up

            # Write out the traceback, because otherwise

            # the catch and throw destroys it

            import traceback

            mutter(traceback.format_exc())

            try:

                if not closed:

                    fout.close()

                self._get_sftp().remove(tmp_abspath)

            except:

                # raise the saved except

                raise e

            # raise the original with its traceback if we can.

            raise

    def _put_non_atomic_helper(self, relpath, writer, mode=None,

                               create_parent_dir=False,

                               dir_mode=None):

        abspath = self._remote_path(relpath)

        # TODO: jam 20060816 paramiko doesn't publicly expose a way to

        #       set the file mode at create time. If it does, use it.

        #       But for now, we just chmod later anyway.

        def _open_and_write_file():

            """Try to open the target file, raise error on failure"""

            fout = None

            try:

                try:

                    fout = self._get_sftp().file(abspath, mode='wb')

                    fout.set_pipelined(True)

                    writer(fout)

                except (paramiko.SSHException, IOError), e:

                    self._translate_io_exception(e, abspath,

                                                 ': unable to open')

                # This is designed to chmod() right before we close.

                # Because we set_pipelined() earlier, theoretically we might 

                # avoid the round trip for fout.close()

                if mode is not None:

                    self._get_sftp().chmod(abspath, mode)

            finally:

                if fout is not None:

                    fout.close()

        if not create_parent_dir:

            _open_and_write_file()

            return

        # Try error handling to create the parent directory if we need to

        try:

            _open_and_write_file()

        except NoSuchFile:

            # Try to create the parent directory, and then go back to

            # writing the file

            parent_dir = os.path.dirname(abspath)

            self._mkdir(parent_dir, dir_mode)

            _open_and_write_file()

    def put_file_non_atomic(self, relpath, f, mode=None,

                            create_parent_dir=False,

                            dir_mode=None):

        """Copy the file-like object into the target location.

        This function is not strictly safe to use. It is only meant to

        be used when you already know that the target does not exist.

        It is not safe, because it will open and truncate the remote

        file. So there may be a time when the file has invalid contents.

        :param relpath: The remote location to put the contents.

        :param f:       File-like object.

        :param mode:    Possible access permissions for new file.

                        None means do not set remote permissions.

        :param create_parent_dir: If we cannot create the target file because

                        the parent directory does not exist, go ahead and

                        create it, and then try again.

        """

        def writer(fout):

            self._pump(f, fout)

        self._put_non_atomic_helper(relpath, writer, mode=mode,

                                    create_parent_dir=create_parent_dir,

                                    dir_mode=dir_mode)

    def put_bytes_non_atomic(self, relpath, bytes, mode=None,

                             create_parent_dir=False,

                             dir_mode=None):

        def writer(fout):

            fout.write(bytes)

        self._put_non_atomic_helper(relpath, writer, mode=mode,

                                    create_parent_dir=create_parent_dir,

                                    dir_mode=dir_mode)

    def iter_files_recursive(self):

        """Walk the relative paths of all files in this transport."""

        queue = list(self.list_dir('.'))

        while queue:

            relpath = queue.pop(0)

            st = self.stat(relpath)

            if stat.S_ISDIR(st.st_mode):

                for i, basename in enumerate(self.list_dir(relpath)):

                    queue.insert(i, relpath+'/'+basename)

            else:

                yield relpath

    def _mkdir(self, abspath, mode=None):

        if mode is None:

            local_mode = 0777

        else:

            local_mode = mode

        try:

            self._get_sftp().mkdir(abspath, local_mode)

            if mode is not None:

                self._get_sftp().chmod(abspath, mode=mode)

        except (paramiko.SSHException, IOError), e:

            self._translate_io_exception(e, abspath, ': unable to mkdir',

                failure_exc=FileExists)

    def mkdir(self, relpath, mode=None):

        """Create a directory at the given path."""

        self._mkdir(self._remote_path(relpath), mode=mode)

    def open_write_stream(self, relpath, mode=None):

        """See Transport.open_write_stream."""

        # initialise the file to zero-length

        # this is three round trips, but we don't use this 

        # api more than once per write_group at the moment so 

        # it is a tolerable overhead. Better would be to truncate

        # the file after opening. RBC 20070805

        self.put_bytes_non_atomic(relpath, "", mode)

        abspath = self._remote_path(relpath)

        # TODO: jam 20060816 paramiko doesn't publicly expose a way to

        #       set the file mode at create time. If it does, use it.

        #       But for now, we just chmod later anyway.

        handle = None

        try:

            handle = self._get_sftp().file(abspath, mode='wb')

            handle.set_pipelined(True)

        except (paramiko.SSHException, IOError), e:

            self._translate_io_exception(e, abspath,

                                         ': unable to open')

        _file_streams[self.abspath(relpath)] = handle

        return FileFileStream(self, relpath, handle)

    def _translate_io_exception(self, e, path, more_info='',

                                failure_exc=PathError):

        """Translate a paramiko or IOError into a friendlier exception.

        :param e: The original exception

        :param path: The path in question when the error is raised

        :param more_info: Extra information that can be included,

                          such as what was going on

        :param failure_exc: Paramiko has the super fun ability to raise completely

                           opaque errors that just set "e.args = ('Failure',)" with

                           no more information.

                           If this parameter is set, it defines the exception 

                           to raise in these cases.

        """

        # paramiko seems to generate detailless errors.

        self._translate_error(e, path, raise_generic=False)

        if getattr(e, 'args', None) is not None:

            if (e.args == ('No such file or directory',) or

                e.args == ('No such file',)):

                raise NoSuchFile(path, str(e) + more_info)

            if (e.args == ('mkdir failed',)):

                raise FileExists(path, str(e) + more_info)

            # strange but true, for the paramiko server.

            if (e.args == ('Failure',)):

                raise failure_exc(path, str(e) + more_info)

            mutter('Raising exception with args %s', e.args)

        if getattr(e, 'errno', None) is not None:

            mutter('Raising exception with errno %s', e.errno)

        raise e

    def append_file(self, relpath, f, mode=None):

        """

        Append the text in the file-like object into the final

        location.

        """

        try:

            path = self._remote_path(relpath)

            fout = self._get_sftp().file(path, 'ab')

            if mode is not None:

                self._get_sftp().chmod(path, mode)

            result = fout.tell()

            self._pump(f, fout)

            return result

        except (IOError, paramiko.SSHException), e:

            self._translate_io_exception(e, relpath, ': unable to append')

    def rename(self, rel_from, rel_to):

        """Rename without special overwriting"""

        try:

            self._get_sftp().rename(self._remote_path(rel_from),

                              self._remote_path(rel_to))

        except (IOError, paramiko.SSHException), e:

            self._translate_io_exception(e, rel_from,

                    ': unable to rename to %r' % (rel_to))

    def _rename_and_overwrite(self, abs_from, abs_to):

        """Do a fancy rename on the remote server.

        Using the implementation provided by osutils.

        """

        try:

            sftp = self._get_sftp()

            fancy_rename(abs_from, abs_to,

                         rename_func=sftp.rename,

                         unlink_func=sftp.remove)

        except (IOError, paramiko.SSHException), e:

            self._translate_io_exception(e, abs_from,

                                         ': unable to rename to %r' % (abs_to))

    def move(self, rel_from, rel_to):

        """Move the item at rel_from to the location at rel_to"""

        path_from = self._remote_path(rel_from)

        path_to = self._remote_path(rel_to)

        self._rename_and_overwrite(path_from, path_to)

    def delete(self, relpath):

        """Delete the item at relpath"""

        path = self._remote_path(relpath)

        try:

            self._get_sftp().remove(path)

        except (IOError, paramiko.SSHException), e:

            self._translate_io_exception(e, path, ': unable to delete')

    def external_url(self):

        """See bzrlib.transport.Transport.external_url."""

        # the external path for SFTP is the base

        return self.base

    def listable(self):

        """Return True if this store supports listing."""

        return True

    def list_dir(self, relpath):

        """

        Return a list of all files at the given location.

        """

        # does anything actually use this?

        # -- Unknown

        # This is at least used by copy_tree for remote upgrades.

        # -- David Allouche 2006-08-11

        path = self._remote_path(relpath)

        try:

            entries = self._get_sftp().listdir(path)

        except (IOError, paramiko.SSHException), e:

            self._translate_io_exception(e, path, ': failed to list_dir')

        return [urlutils.escape(entry) for entry in entries]