Viewing file:  filepath.py (32.17 KB) -rw-r--r--
Select action/file-type:
 (+) |  (+) |  (+) | Code (+) | Session (+) |  (+) | SDB (+) |  (+) |  (+) |  (+) |  (+) |  (+) |
# -*- test-case-name: twisted.test.test_paths -*-
# Copyright (c) 2001-2010 Twisted Matrix Laboratories.
# See LICENSE for details.
"""
Object-oriented filesystem path representation.
"""
import os
import errno
import random
import base64
from os.path import isabs, exists, normpath, abspath, splitext
from os.path import basename, dirname
from os.path import join as joinpath
from os import sep as slash
from os import listdir, utime, stat
from stat import S_ISREG, S_ISDIR
# Please keep this as light as possible on other Twisted imports; many, many
# things import this module, and it would be good if it could easily be
# modified for inclusion in the standard library. --glyph
from twisted.python.runtime import platform
from twisted.python.hashlib import sha1
from twisted.python.win32 import ERROR_FILE_NOT_FOUND, ERROR_PATH_NOT_FOUND
from twisted.python.win32 import ERROR_INVALID_NAME, ERROR_DIRECTORY, O_BINARY
from twisted.python.win32 import WindowsError
_CREATE_FLAGS = (os.O_EXCL |
os.O_CREAT |
os.O_RDWR |
O_BINARY)
def _stub_islink(path):
"""
Always return 'false' if the operating system does not support symlinks.
@param path: a path string.
@type path: L{str}
@return: false
"""
return False
def _stub_urandom(n):
"""
Provide random data in versions of Python prior to 2.4. This is an
effectively compatible replacement for 'os.urandom'.
@type n: L{int}
@param n: the number of bytes of data to return
@return: C{n} bytes of random data.
@rtype: str
"""
randomData = [random.randrange(256) for n in xrange(n)]
return ''.join(map(chr, randomData))
def _stub_armor(s):
"""
ASCII-armor for random data. This uses a hex encoding, although we will
prefer url-safe base64 encoding for features in this module if it is
available.
"""
return s.encode('hex')
islink = getattr(os.path, 'islink', _stub_islink)
randomBytes = getattr(os, 'urandom', _stub_urandom)
armor = getattr(base64, 'urlsafe_b64encode', _stub_armor)
class InsecurePath(Exception):
"""
Error that is raised when the path provided to FilePath is invalid.
"""
class LinkError(Exception):
"""
An error with symlinks - either that there are cyclical symlinks or that
symlink are not supported on this platform.
"""
class UnlistableError(OSError):
"""
An exception which is used to distinguish between errors which mean 'this
is not a directory you can list' and other, more catastrophic errors.
This error will try to look as much like the original error as possible,
while still being catchable as an independent type.
@ivar originalException: the actual original exception instance, either an
L{OSError} or a L{WindowsError}.
"""
def __init__(self, originalException):
"""
Create an UnlistableError exception.
@param originalException: an instance of OSError.
"""
self.__dict__.update(originalException.__dict__)
self.originalException = originalException
class _WindowsUnlistableError(UnlistableError, WindowsError):
"""
This exception is raised on Windows, for compatibility with previous
releases of FilePath where unportable programs may have done "except
WindowsError:" around a call to children().
It is private because all application code may portably catch
L{UnlistableError} instead.
"""
def _secureEnoughString():
"""
Create a pseudorandom, 16-character string for use in secure filenames.
"""
return armor(sha1(randomBytes(64)).digest())[:16]
class _PathHelper:
"""
Abstract helper class also used by ZipPath; implements certain utility
methods.
"""
def getContent(self):
fp = self.open()
try:
return fp.read()
finally:
fp.close()
def parents(self):
"""
@return: an iterator of all the ancestors of this path, from the most
recent (its immediate parent) to the root of its filesystem.
"""
path = self
parent = path.parent()
# root.parent() == root, so this means "are we the root"
while path != parent:
yield parent
path = parent
parent = parent.parent()
def children(self):
"""
List the children of this path object.
@raise OSError: If an error occurs while listing the directory. If the
error is 'serious', meaning that the operation failed due to an access
violation, exhaustion of some kind of resource (file descriptors or
memory), OSError or a platform-specific variant will be raised.
@raise UnlistableError: If the inability to list the directory is due
to this path not existing or not being a directory, the more specific
OSError subclass L{UnlistableError} is raised instead.
@return: an iterable of all currently-existing children of this object
accessible with L{_PathHelper.child}.
"""
try:
subnames = self.listdir()
except WindowsError, winErrObj:
# WindowsError is an OSError subclass, so if not for this clause
# the OSError clause below would be handling these. Windows error
# codes aren't the same as POSIX error codes, so we need to handle
# them differently.
# Under Python 2.5 on Windows, WindowsError has a winerror
# attribute and an errno attribute. The winerror attribute is
# bound to the Windows error code while the errno attribute is
# bound to a translation of that code to a perhaps equivalent POSIX
# error number.
# Under Python 2.4 on Windows, WindowsError only has an errno
# attribute. It is bound to the Windows error code.
# For simplicity of code and to keep the number of paths through
# this suite minimal, we grab the Windows error code under either
# version.
# Furthermore, attempting to use os.listdir on a non-existent path
# in Python 2.4 will result in a Windows error code of
# ERROR_PATH_NOT_FOUND. However, in Python 2.5,
# ERROR_FILE_NOT_FOUND results instead. -exarkun
winerror = getattr(winErrObj, 'winerror', winErrObj.errno)
if winerror not in (ERROR_PATH_NOT_FOUND,
ERROR_FILE_NOT_FOUND,
ERROR_INVALID_NAME,
ERROR_DIRECTORY):
raise
raise _WindowsUnlistableError(winErrObj)
except OSError, ose:
if ose.errno not in (errno.ENOENT, errno.ENOTDIR):
# Other possible errors here, according to linux manpages:
# EACCES, EMIFLE, ENFILE, ENOMEM. None of these seem like the
# sort of thing which should be handled normally. -glyph
raise
raise UnlistableError(ose)
return map(self.child, subnames)
def walk(self, descend=None):
"""
Yield myself, then each of my children, and each of those children's
children in turn. The optional argument C{descend} is a predicate that
takes a FilePath, and determines whether or not that FilePath is
traversed/descended into. It will be called with each path for which
C{isdir} returns C{True}. If C{descend} is not specified, all
directories will be traversed (including symbolic links which refer to
directories).
@param descend: A one-argument callable that will return True for
FilePaths that should be traversed, False otherwise.
@return: a generator yielding FilePath-like objects.
"""
yield self
if self.isdir():
for c in self.children():
# we should first see if it's what we want, then we
# can walk through the directory
if (descend is None or descend(c)):
for subc in c.walk(descend):
if os.path.realpath(self.path).startswith(
os.path.realpath(subc.path)):
raise LinkError("Cycle in file graph.")
yield subc
else:
yield c
def sibling(self, path):
return self.parent().child(path)
def segmentsFrom(self, ancestor):
"""
Return a list of segments between a child and its ancestor.
For example, in the case of a path X representing /a/b/c/d and a path Y
representing /a/b, C{Y.segmentsFrom(X)} will return C{['c',
'd']}.
@param ancestor: an instance of the same class as self, ostensibly an
ancestor of self.
@raise: ValueError if the 'ancestor' parameter is not actually an
ancestor, i.e. a path for /x/y/z is passed as an ancestor for /a/b/c/d.
@return: a list of strs
"""
# this might be an unnecessarily inefficient implementation but it will
# work on win32 and for zipfiles; later I will deterimine if the
# obvious fast implemenation does the right thing too
f = self
p = f.parent()
segments = []
while f != ancestor and p != f:
segments[0:0] = [f.basename()]
f = p
p = p.parent()
if f == ancestor and segments:
return segments
raise ValueError("%r not parent of %r" % (ancestor, self))
# new in 8.0
def __hash__(self):
"""
Hash the same as another FilePath with the same path as mine.
"""
return hash((self.__class__, self.path))
# pending deprecation in 8.0
def getmtime(self):
"""
Deprecated. Use getModificationTime instead.
"""
return int(self.getModificationTime())
def getatime(self):
"""
Deprecated. Use getAccessTime instead.
"""
return int(self.getAccessTime())
def getctime(self):
"""
Deprecated. Use getStatusChangeTime instead.
"""
return int(self.getStatusChangeTime())
class FilePath(_PathHelper):
"""
I am a path on the filesystem that only permits 'downwards' access.
Instantiate me with a pathname (for example,
FilePath('/home/myuser/public_html')) and I will attempt to only provide
access to files which reside inside that path. I may be a path to a file,
a directory, or a file which does not exist.
The correct way to use me is to instantiate me, and then do ALL filesystem
access through me. In other words, do not import the 'os' module; if you
need to open a file, call my 'open' method. If you need to list a
directory, call my 'path' method.
Even if you pass me a relative path, I will convert that to an absolute
path internally.
Note: although time-related methods do return floating-point results, they
may still be only second resolution depending on the platform and the last
value passed to L{os.stat_float_times}. If you want greater-than-second
precision, call C{os.stat_float_times(True)}, or use Python 2.5.
Greater-than-second precision is only available in Windows on Python2.5 and
later.
@type alwaysCreate: C{bool}
@ivar alwaysCreate: When opening this file, only succeed if the file does
not already exist.
@type path: C{str}
@ivar path: The path from which 'downward' traversal is permitted.
@ivar statinfo: The currently cached status information about the file on
the filesystem that this L{FilePath} points to. This attribute is
C{None} if the file is in an indeterminate state (either this
L{FilePath} has not yet had cause to call C{stat()} yet or
L{FilePath.changed} indicated that new information is required), 0 if
C{stat()} was called and returned an error (i.e. the path did not exist
when C{stat()} was called), or a C{stat_result} object that describes
the last known status of the underlying file (or directory, as the case
may be). Trust me when I tell you that you do not want to use this
attribute. Instead, use the methods on L{FilePath} which give you
information about it, like C{getsize()}, C{isdir()},
C{getModificationTime()}, and so on.
@type statinfo: C{int} or L{types.NoneType} or L{os.stat_result}
"""
statinfo = None
path = None
def __init__(self, path, alwaysCreate=False):
self.path = abspath(path)
self.alwaysCreate = alwaysCreate
def __getstate__(self):
d = self.__dict__.copy()
if d.has_key('statinfo'):
del d['statinfo']
return d
def child(self, path):
if platform.isWindows() and path.count(":"):
# Catch paths like C:blah that don't have a slash
raise InsecurePath("%r contains a colon." % (path,))
norm = normpath(path)
if slash in norm:
raise InsecurePath("%r contains one or more directory separators" % (path,))
newpath = abspath(joinpath(self.path, norm))
if not newpath.startswith(self.path):
raise InsecurePath("%r is not a child of %s" % (newpath, self.path))
return self.clonePath(newpath)
def preauthChild(self, path):
"""
Use me if `path' might have slashes in it, but you know they're safe.
(NOT slashes at the beginning. It still needs to be a _child_).
"""
newpath = abspath(joinpath(self.path, normpath(path)))
if not newpath.startswith(self.path):
raise InsecurePath("%s is not a child of %s" % (newpath, self.path))
return self.clonePath(newpath)
def childSearchPreauth(self, *paths):
"""Return my first existing child with a name in 'paths'.
paths is expected to be a list of *pre-secured* path fragments; in most
cases this will be specified by a system administrator and not an
arbitrary user.
If no appropriately-named children exist, this will return None.
"""
p = self.path
for child in paths:
jp = joinpath(p, child)
if exists(jp):
return self.clonePath(jp)
def siblingExtensionSearch(self, *exts):
"""Attempt to return a path with my name, given multiple possible
extensions.
Each extension in exts will be tested and the first path which exists
will be returned. If no path exists, None will be returned. If '' is
in exts, then if the file referred to by this path exists, 'self' will
be returned.
The extension '*' has a magic meaning, which means "any path that
begins with self.path+'.' is acceptable".
"""
p = self.path
for ext in exts:
if not ext and self.exists():
return self
if ext == '*':
basedot = basename(p)+'.'
for fn in listdir(dirname(p)):
if fn.startswith(basedot):
return self.clonePath(joinpath(dirname(p), fn))
p2 = p + ext
if exists(p2):
return self.clonePath(p2)
def realpath(self):
"""
Returns the absolute target as a FilePath if self is a link, self
otherwise. The absolute link is the ultimate file or directory the
link refers to (for instance, if the link refers to another link, and
another...). If the filesystem does not support symlinks, or
if the link is cyclical, raises a LinkError.
Behaves like L{os.path.realpath} in that it does not resolve link
names in the middle (ex. /x/y/z, y is a link to w - realpath on z
will return /x/y/z, not /x/w/z).
@return: FilePath of the target path
@raises LinkError: if links are not supported or links are cyclical.
"""
if self.islink():
result = os.path.realpath(self.path)
if result == self.path:
raise LinkError("Cyclical link - will loop forever")
return self.clonePath(result)
return self
def siblingExtension(self, ext):
return self.clonePath(self.path+ext)
def linkTo(self, linkFilePath):
"""
Creates a symlink to self to at the path in the L{FilePath}
C{linkFilePath}. Only works on posix systems due to its dependence on
C{os.symlink}. Propagates C{OSError}s up from C{os.symlink} if
C{linkFilePath.parent()} does not exist, or C{linkFilePath} already
exists.
@param linkFilePath: a FilePath representing the link to be created
@type linkFilePath: L{FilePath}
"""
os.symlink(self.path, linkFilePath.path)
def open(self, mode='r'):
"""
Open this file using C{mode} or for writing if C{alwaysCreate} is
C{True}.
In all cases the file is opened in binary mode, so it is not necessary
to include C{b} in C{mode}.
@param mode: The mode to open the file in. Default is C{r}.
@type mode: C{str}
@raises AssertionError: If C{a} is included in the mode and
C{alwaysCreate} is C{True}.
@rtype: C{file}
@return: An open C{file} object.
"""
if self.alwaysCreate:
assert 'a' not in mode, ("Appending not supported when "
"alwaysCreate == True")
return self.create()
# This hack is necessary because of a bug in Python 2.7 on Windows:
# http://bugs.python.org/issue7686
mode = mode.replace('b', '')
return open(self.path, mode + 'b')
# stat methods below
def restat(self, reraise=True):
"""
Re-calculate cached effects of 'stat'. To refresh information on this path
after you know the filesystem may have changed, call this method.
@param reraise: a boolean. If true, re-raise exceptions from
L{os.stat}; otherwise, mark this path as not existing, and remove any
cached stat information.
"""
try:
self.statinfo = stat(self.path)
except OSError:
self.statinfo = 0
if reraise:
raise
def changed(self):
"""
Clear any cached information about the state of this path on disk.
@since: 10.1.0
"""
self.statinfo = None
def chmod(self, mode):
"""
Changes the permissions on self, if possible. Propagates errors from
C{os.chmod} up.
@param mode: integer representing the new permissions desired (same as
the command line chmod)
@type mode: C{int}
"""
os.chmod(self.path, mode)
def getsize(self):
st = self.statinfo
if not st:
self.restat()
st = self.statinfo
return st.st_size
def getModificationTime(self):
"""
Retrieve the time of last access from this file.
@return: a number of seconds from the epoch.
@rtype: float
"""
st = self.statinfo
if not st:
self.restat()
st = self.statinfo
return float(st.st_mtime)
def getStatusChangeTime(self):
"""
Retrieve the time of the last status change for this file.
@return: a number of seconds from the epoch.
@rtype: float
"""
st = self.statinfo
if not st:
self.restat()
st = self.statinfo
return float(st.st_ctime)
def getAccessTime(self):
"""
Retrieve the time that this file was last accessed.
@return: a number of seconds from the epoch.
@rtype: float
"""
st = self.statinfo
if not st:
self.restat()
st = self.statinfo
return float(st.st_atime)
def exists(self):
"""
Check if the C{path} exists.
@return: C{True} if the stats of C{path} can be retrieved successfully,
C{False} in the other cases.
@rtype: C{bool}
"""
if self.statinfo:
return True
else:
self.restat(False)
if self.statinfo:
return True
else:
return False
def isdir(self):
st = self.statinfo
if not st:
self.restat(False)
st = self.statinfo
if not st:
return False
return S_ISDIR(st.st_mode)
def isfile(self):
st = self.statinfo
if not st:
self.restat(False)
st = self.statinfo
if not st:
return False
return S_ISREG(st.st_mode)
def islink(self):
# We can't use cached stat results here, because that is the stat of
# the destination - (see #1773) which in *every case* but this one is
# the right thing to use. We could call lstat here and use that, but
# it seems unlikely we'd actually save any work that way. -glyph
return islink(self.path)
def isabs(self):
return isabs(self.path)
def listdir(self):
return listdir(self.path)
def splitext(self):
return splitext(self.path)
def __repr__(self):
return 'FilePath(%r)' % (self.path,)
def touch(self):
try:
self.open('a').close()
except IOError:
pass
utime(self.path, None)
def remove(self):
"""
Removes the file or directory that is represented by self. If
C{self.path} is a directory, recursively remove all its children
before removing the directory. If it's a file or link, just delete
it.
"""
if self.isdir() and not self.islink():
for child in self.children():
child.remove()
os.rmdir(self.path)
else:
os.remove(self.path)
self.changed()
def makedirs(self):
"""
Create all directories not yet existing in C{path} segments, using
C{os.makedirs}.
"""
return os.makedirs(self.path)
def globChildren(self, pattern):
"""
Assuming I am representing a directory, return a list of
FilePaths representing my children that match the given
pattern.
"""
import glob
path = self.path[-1] == '/' and self.path + pattern or slash.join([self.path, pattern])
return map(self.clonePath, glob.glob(path))
def basename(self):
return basename(self.path)
def dirname(self):
return dirname(self.path)
def parent(self):
return self.clonePath(self.dirname())
def setContent(self, content, ext='.new'):
"""
Replace the file at this path with a new file that contains the given
bytes, trying to avoid data-loss in the meanwhile.
On UNIX-like platforms, this method does its best to ensure that by the
time this method returns, either the old contents I{or} the new contents
of the file will be present at this path for subsequent readers
regardless of premature device removal, program crash, or power loss,
making the following assumptions:
- your filesystem is journaled (i.e. your filesystem will not
I{itself} lose data due to power loss)
- your filesystem's C{rename()} is atomic
- your filesystem will not discard new data while preserving new
metadata (see U{http://mjg59.livejournal.com/108257.html} for more
detail)
On most versions of Windows there is no atomic C{rename()} (see
U{http://bit.ly/win32-overwrite} for more information), so this method
is slightly less helpful. There is a small window where the file at
this path may be deleted before the new file is moved to replace it:
however, the new file will be fully written and flushed beforehand so in
the unlikely event that there is a crash at that point, it should be
possible for the user to manually recover the new version of their data.
In the future, Twisted will support atomic file moves on those versions
of Windows which I{do} support them: see U{Twisted ticket
3004<http://twistedmatrix.com/trac/ticket/3004>}.
This method should be safe for use by multiple concurrent processes, but
note that it is not easy to predict which process's contents will
ultimately end up on disk if they invoke this method at close to the
same time.
@param content: The desired contents of the file at this path.
@type content: L{str}
@param ext: An extension to append to the temporary filename used to
store the bytes while they are being written. This can be used to
make sure that temporary files can be identified by their suffix,
for cleanup in case of crashes.
@type ext: C{str}
"""
sib = self.temporarySibling(ext)
f = sib.open('w')
try:
f.write(content)
finally:
f.close()
if platform.isWindows() and exists(self.path):
os.unlink(self.path)
os.rename(sib.path, self.path)
# new in 2.2.0
def __cmp__(self, other):
if not isinstance(other, FilePath):
return NotImplemented
return cmp(self.path, other.path)
def createDirectory(self):
os.mkdir(self.path)
def requireCreate(self, val=1):
self.alwaysCreate = val
def create(self):
"""Exclusively create a file, only if this file previously did not exist.
"""
fdint = os.open(self.path, _CREATE_FLAGS)
# XXX TODO: 'name' attribute of returned files is not mutable or
# settable via fdopen, so this file is slighly less functional than the
# one returned from 'open' by default. send a patch to Python...
return os.fdopen(fdint, 'w+b')
def temporarySibling(self, extension=""):
"""
Construct a path referring to a sibling of this path.
The resulting path will be unpredictable, so that other subprocesses
should neither accidentally attempt to refer to the same path before it
is created, nor they should other processes be able to guess its name in
advance.
@param extension: A suffix to append to the created filename. (Note
that if you want an extension with a '.' you must include the '.'
yourself.)
@type extension: C{str}
@return: a path object with the given extension suffix, C{alwaysCreate}
set to True.
@rtype: L{FilePath}
"""
sib = self.sibling(_secureEnoughString() + self.basename() + extension)
sib.requireCreate()
return sib
_chunkSize = 2 ** 2 ** 2 ** 2
def copyTo(self, destination, followLinks=True):
"""
Copies self to destination.
If self is a directory, this method copies its children (but not
itself) recursively to destination - if destination does not exist as a
directory, this method creates it. If destination is a file, an
IOError will be raised.
If self is a file, this method copies it to destination. If
destination is a file, this method overwrites it. If destination is a
directory, an IOError will be raised.
If self is a link (and followLinks is False), self will be copied
over as a new symlink with the same target as returned by os.readlink.
That means that if it is absolute, both the old and new symlink will
link to the same thing. If it's relative, then perhaps not (and
it's also possible that this relative link will be broken).
File/directory permissions and ownership will NOT be copied over.
If followLinks is True, symlinks are followed so that they're treated
as their targets. In other words, if self is a link, the link's target
will be copied. If destination is a link, self will be copied to the
destination's target (the actual destination will be destination's
target). Symlinks under self (if self is a directory) will be
followed and its target's children be copied recursively.
If followLinks is False, symlinks will be copied over as symlinks.
@param destination: the destination (a FilePath) to which self
should be copied
@param followLinks: whether symlinks in self should be treated as links
or as their targets
"""
if self.islink() and not followLinks:
os.symlink(os.readlink(self.path), destination.path)
return
# XXX TODO: *thorough* audit and documentation of the exact desired
# semantics of this code. Right now the behavior of existent
# destination symlinks is convenient, and quite possibly correct, but
# its security properties need to be explained.
if self.isdir():
if not destination.exists():
destination.createDirectory()
for child in self.children():
destChild = destination.child(child.basename())
child.copyTo(destChild, followLinks)
elif self.isfile():
writefile = destination.open('w')
try:
readfile = self.open()
try:
while 1:
# XXX TODO: optionally use os.open, os.read and O_DIRECT
# and use os.fstatvfs to determine chunk sizes and make
# *****sure**** copy is page-atomic; the following is
# good enough for 99.9% of everybody and won't take a
# week to audit though.
chunk = readfile.read(self._chunkSize)
writefile.write(chunk)
if len(chunk) < self._chunkSize:
break
finally:
readfile.close()
finally:
writefile.close()
else:
# If you see the following message because you want to copy
# symlinks, fifos, block devices, character devices, or unix
# sockets, please feel free to add support to do sensible things in
# reaction to those types!
raise NotImplementedError(
"Only copying of files and directories supported")
def moveTo(self, destination, followLinks=True):
"""
Move self to destination - basically renaming self to whatever
destination is named. If destination is an already-existing directory,
moves all children to destination if destination is empty. If
destination is a non-empty directory, or destination is a file, an
OSError will be raised.
If moving between filesystems, self needs to be copied, and everything
that applies to copyTo applies to moveTo.
@param destination: the destination (a FilePath) to which self
should be copied
@param followLinks: whether symlinks in self should be treated as links
or as their targets (only applicable when moving between
filesystems)
"""
try:
os.rename(self.path, destination.path)
except OSError, ose:
if ose.errno == errno.EXDEV:
# man 2 rename, ubuntu linux 5.10 "breezy":
# oldpath and newpath are not on the same mounted filesystem.
# (Linux permits a filesystem to be mounted at multiple
# points, but rename(2) does not work across different mount
# points, even if the same filesystem is mounted on both.)
# that means it's time to copy trees of directories!
secsib = destination.temporarySibling()
self.copyTo(secsib, followLinks) # slow
secsib.moveTo(destination, followLinks) # visible
# done creating new stuff. let's clean me up.
mysecsib = self.temporarySibling()
self.moveTo(mysecsib, followLinks) # visible
mysecsib.remove() # slow
else:
raise
else:
self.changed()
destination.changed()
FilePath.clonePath = FilePath
|