Ticket #2070: 5343_1_streaming_file_upload.diff

django/http/multipartparser.py

@@ +1 @@
+"""
+MultiPart parsing for file uploads.
+If both a progress id is sent (either through ``X-Progress-ID``
+header or ``progress_id`` GET) and ``FILE_UPLOAD_DIR`` is set
+in the settings, then the file progress will be tracked using
+``request.file_progress``.
+
+To use this feature, consider creating a middleware with an appropriate
+``process_request``::
+
+    class FileProgressTrack(object):
+        def __get__(self, request, HttpRequest):
+            progress_id = request.META['UPLOAD_PROGRESS_ID']
+            status = # get progress from progress_id here
+
+            return status
+
+        def __set__(self, request, new_value):
+            progress_id = request.META['UPLOAD_PROGRESS_ID']
+
+            # set the progress using progress_id here.
+
+    # example middleware
+    class FileProgressExample(object):
+        def process_request(self, request):
+            request.__class__.file_progress = FileProgressTrack()
+
+
+
+"""
+
+__all__ = ['MultiPartParserError','MultiPartParser']
+
+
+from django.utils.datastructures import MultiValueDict
+import os
+
+try:
+    from cStringIO import StringIO
+except ImportError:
+    from StringIO import StringIO
+
+
+class MultiPartParserError(Exception):
+    def __init__(self, message):
+        self.message = message
+    def __str__(self):
+        return repr(self.message)
+
+class MultiPartParser(object):
+    """
+    A rfc2388 multipart/form-data parser.
+    
+    parse() reads the input stream in chunk_size chunks and returns a
+    tuple of (POST MultiValueDict, FILES MultiValueDict). If
+    file_upload_dir is defined files will be streamed to temporary
+    files in the specified directory.
+
+    The FILES dictionary will have 'filename', 'content-type',
+    'content' and 'content-length' entries. For streamed files it will
+    also have 'tmpfilename' and 'tmpfile'. The 'content' entry will
+    only be read from disk when referenced for streamed files.
+
+    If the X-Progress-ID is sent (in one of many formats), then
+    object.file_progress will be given a dictionary of the progress.
+    """
+    def __init__(self, headers, input, request, file_upload_dir=None, streaming_min_post_size=None, chunk_size=1024*64):
+        try:
+            content_length = int(headers['Content-Length'])
+        except:
+            raise MultiPartParserError('Invalid Content-Length: %s' % headers.get('Content-Length'))
+
+        content_type = headers.get('Content-Type')
+
+        if not content_type or not content_type.startswith('multipart/'):
+            raise MultiPartParserError('Invalid Content-Type: %s' % content_type)
+            
+        ctype, opts = self.parse_header(content_type)
+        boundary = opts.get('boundary')
+        from cgi import valid_boundary
+        if not boundary or not valid_boundary(boundary):
+            raise MultiPartParserError('Invalid boundary in multipart form: %s' % boundary)
+
+        progress_id = request.META['UPLOAD_PROGRESS_ID']
+
+        self._track_progress = file_upload_dir and progress_id # whether or not to track progress
+        self._boundary = '--' + boundary
+        self._input = input
+        self._size = content_length
+        self._received = 0
+        self._file_upload_dir = file_upload_dir
+        self._chunk_size = chunk_size
+        self._state = 'PREAMBLE'
+        self._partial = ''
+        self._post = MultiValueDict()
+        self._files = MultiValueDict()
+        self._request = request
+
+        if streaming_min_post_size is not None and content_length < streaming_min_post_size:
+            self._file_upload_dir = None # disable file streaming for small request
+        elif self._track_progress:
+            request.file_progress = {'state': 'starting'}
+
+        try:
+            # Use mx fast string search if available.
+            from mx.TextTools import FS
+            self._fs = FS(self._boundary)
+        except ImportError:
+            self._fs = None
+
+    def parse(self):
+        try:
+            self._parse()
+        finally:
+            if self._track_progress:
+                self._request.file_progress = {'state': 'done'}
+        return self._post, self._files
+
+    def _parse(self):
+        size = self._size
+
+        try:
+            while size > 0:
+                n = self._read(self._input, min(self._chunk_size, size))
+                if not n:
+                    break
+                size -= n
+        except:
+            # consume any remaining data so we dont generate a "Connection Reset" error
+            size = self._size - self._received
+            while size > 0:
+                data = self._input.read(min(self._chunk_size, size))
+                size -= len(data)
+            raise
+
+    def _find_boundary(self, data, start, stop):
+        """
+        Find the next boundary and return the end of current part
+        and start of next part.
+        """
+        if self._fs:
+            boundary = self._fs.find(data, start, stop)
+        else:
+            boundary = data.find(self._boundary, start, stop)
+        if boundary >= 0:
+            end = boundary
+            next = boundary + len(self._boundary)
+
+            # backup over CRLF
+            if end > 0 and data[end-1] == '\n': end -= 1
+            if end > 0 and data[end-1] == '\r': end -= 1
+            # skip over --CRLF
+            if next < stop and data[next] == '-': next += 1
+            if next < stop and data[next] == '-': next += 1
+            if next < stop and data[next] == '\r': next += 1
+            if next < stop and data[next] == '\n': next += 1
+
+            return True, end, next
+        else:
+            return False, stop, stop
+
+    class TemporaryFile(object):
+        "A temporary file that tries to delete itself when garbage collected."
+        def __init__(self, dir):
+            import tempfile
+            (fd, name) = tempfile.mkstemp(suffix='.upload', dir=dir)
+            self.file = os.fdopen(fd, 'w+b')
+            self.name = name
+
+        def __getattr__(self, name):
+            a = getattr(self.__dict__['file'], name)
+            if type(a) != type(0):
+                setattr(self, name, a)
+            return a
+
+        def __del__(self):
+            try:
+                os.unlink(self.name)
+            except OSError:
+                pass
+
+    class LazyContent(dict):
+        """
+        A lazy FILES dictionary entry that reads the contents from
+        tmpfile only when referenced.
+        """
+        def __init__(self, data):
+            dict.__init__(self, data)
+
+        def __getitem__(self, key):
+            if key == 'content' and not self.has_key(key):
+                self['tmpfile'].seek(0)
+                self['content'] = self['tmpfile'].read()
+            return dict.__getitem__(self, key)
+
+    def _read(self, input, size):
+        data = input.read(size)
+
+        if not data:
+            return 0
+
+        read_size = len(data)
+        self._received += read_size
+
+        if self._partial:
+            data = self._partial + data
+
+        start = 0
+        stop = len(data)
+
+        while start < stop:
+            boundary, end, next = self._find_boundary(data, start, stop)
+
+            if not boundary and read_size:
+                # make sure we dont treat a partial boundary (and its separators) as data
+                stop -= len(self._boundary) + 16
+                end = next = stop
+                if end <= start:
+                    break # need more data
+
+            if self._state == 'PREAMBLE':
+                # Preamble, just ignore it
+                self._state = 'HEADER'
+
+            elif self._state == 'HEADER':
+                # Beginning of header, look for end of header and parse it if found.
+
+                header_end = data.find('\r\n\r\n', start, stop)
+                if header_end == -1:
+                    break # need more data
+
+                header = data[start:header_end]
+
+                self._fieldname = None
+                self._filename = None
+                self._content_type = None
+
+                for line in header.split('\r\n'):
+                    ctype, opts = self.parse_header(line)
+                    if ctype == 'content-disposition: form-data':
+                        self._fieldname = opts.get('name')
+                        self._filename = opts.get('filename')
+                    elif ctype.startswith('content-type: '):
+                        self._content_type = ctype[14:]
+
+                if self._filename is not None:
+                    # cleanup filename from IE full paths:
+                    self._filename = self._filename[self._filename.rfind("\\")+1:].strip()
+
+                    if self._filename: # ignore files without filenames
+                        if self._file_upload_dir:
+#                            assert False, 'hmm'
+                            
+                            try:
+                                self._file = self.TemporaryFile(dir=self._file_upload_dir)
+                            except OSError, IOError:
+                                raise MultiPartParserError("Failed to create temporary file. Error was %s" % e)
+                        else:
+                            self._file = StringIO()
+                    else:
+                        self._file = None
+                    self._filesize = 0
+                    self._state = 'FILE'
+                else:
+                    self._field = StringIO()
+                    self._state = 'FIELD'
+                next = header_end + 4
+
+            elif self._state == 'FIELD':
+                # In a field, collect data until a boundary is found.
+
+                self._field.write(data[start:end])
+                if boundary:
+                    if self._fieldname:
+                        self._post.appendlist(self._fieldname, self._field.getvalue())
+                    self._field.close()
+                    self._state = 'HEADER'
+
+            elif self._state == 'FILE':
+                # In a file, collect data until a boundary is found.
+
+                if self._file:
+                    try:
+                        self._file.write(data[start:end])
+                    except IOError, e:
+                        raise MultiPartParserError("Failed to write to temporary file.")
+                    self._filesize += end-start
+
+                    if self._track_progress:
+                        self._request.file_progress = {'received': self._received,
+                                                       'size':     self._size,
+                                                       'state':    'uploading'}
+
+                if boundary:
+                    if self._file:
+                        if self._file_upload_dir:
+                            self._file.seek(0)
+                            file = self.LazyContent({
+                                'filename': self._filename,
+                                'content-type':  self._content_type,
+                                # 'content': is read on demand
+                                'content-length': self._filesize,
+                                'tmpfilename': self._file.name,
+                                'tmpfile': self._file
+                            })
+                        else:
+                            file = {
+                                'filename': self._filename,
+                                'content-type':  self._content_type,
+                                'content': self._file.getvalue(),
+                                'content-length': self._filesize
+                            }
+                            self._file.close()
+
+                        self._files.appendlist(self._fieldname, file)
+
+                    self._state = 'HEADER'
+
+            start = next
+
+        self._partial = data[start:]
+
+        return read_size
+
+    def parse_header(self, line):
+        from cgi import parse_header
+        return parse_header(line)

django/http/__init__.py

@@ -1 @@
-import os
+import os, pickle
 from Cookie import SimpleCookie
 from pprint import pformat
 from urllib import urlencode, quote
 from django.utils.datastructures import MultiValueDict
+from django.http.multipartparser import MultiPartParser, MultiPartParserError
+import re
 
+upload_id_re = re.compile(r'^[a-fA-F0-9]{32}$') # file progress id Regular expression
+
 RESERVED_CHARS="!*'();:@&=+$,/?%#[]"
 
 try:
@@ -17 +21 @@
 
 class HttpRequest(object):
     "A basic HTTP request"
+
     def __init__(self):
         self.GET, self.POST, self.COOKIES, self.META, self.FILES = {}, {}, {}, {}, {}
         self.path = ''
@@ -42 +47 @@
     def is_secure(self):
         return os.environ.get("HTTPS") == "on"
 
-def parse_file_upload(header_dict, post_data):
-    "Returns a tuple of (POST MultiValueDict, FILES MultiValueDict)"
-    import email, email.Message
-    from cgi import parse_header
-    raw_message = '\r\n'.join(['%s:%s' % pair for pair in header_dict.items()])
-    raw_message += '\r\n\r\n' + post_data
-    msg = email.message_from_string(raw_message)
-    POST = MultiValueDict()
-    FILES = MultiValueDict()
-    for submessage in msg.get_payload():
-        if submessage and isinstance(submessage, email.Message.Message):
-            name_dict = parse_header(submessage['Content-Disposition'])[1]
-            # name_dict is something like {'name': 'file', 'filename': 'test.txt'} for file uploads
-            # or {'name': 'blah'} for POST fields
-            # We assume all uploaded files have a 'filename' set.
-            if 'filename' in name_dict:
-                assert type([]) != type(submessage.get_payload()), "Nested MIME messages are not supported"
-                if not name_dict['filename'].strip():
-                    continue
-                # IE submits the full path, so trim everything but the basename.
-                # (We can't use os.path.basename because it expects Linux paths.)
-                filename = name_dict['filename'][name_dict['filename'].rfind("\\")+1:]
-                FILES.appendlist(name_dict['name'], {
-                    'filename': filename,
-                    'content-type': 'Content-Type' in submessage and submessage['Content-Type'] or None,
-                    'content': submessage.get_payload(),
-                })
-            else:
-                POST.appendlist(name_dict['name'], submessage.get_payload())
-    return POST, FILES
+    def _get_file_progress(self):
+        return {}
 
+    def _set_file_progress(self,value):
+        pass
+
+    def _del_file_progress(self):
+        pass
+
+    file_progress = property(_get_file_progress,
+                             _set_file_progress,
+                             _del_file_progress)
+
+    def _get_file_progress_from_args(self, headers, get, querystring):
+        """
+        This parses the request for a file progress_id value.
+        Note that there are two distinct ways of getting the progress
+        ID -- header and GET. One is used primarily to attach via JavaScript
+        to the end of an HTML form action while the other is used for AJAX
+        communication.
+
+        All progress IDs must be valid 32-digit hexadecimal numbers.
+        """
+        if 'X-Progress-ID' in headers:
+            progress_id = headers['X-Upload-ID']
+        elif 'progress_id' in get:
+            progress_id = get['progress_id']
+        else:
+            return None
+
+        if not self.upload_id_re.match(progress_id):
+            return None
+
+        return progress_id
+
+def parse_file_upload(headers, input, request):
+    from django.conf import settings
+
+    # Only stream files to disk if FILE_STREAMING_DIR is set
+    file_upload_dir = settings.FILE_UPLOAD_DIR
+    streaming_min_post_size = settings.STREAMING_MIN_POST_SIZE
+
+    try:
+        parser = MultiPartParser(headers, input, request, file_upload_dir, streaming_min_post_size)
+        return parser.parse()
+    except MultiPartParserError, e:
+        return MultiValueDict({ '_file_upload_error': [e.message] }), {}
+
+
 class QueryDict(MultiValueDict):
     """A specialized MultiValueDict that takes a query string when initialized.
     This is immutable unless you create a copy of it."""

django/oldforms/__init__.py

@@ -666 +666 @@
         self.validator_list = [self.isNonEmptyFile] + validator_list
 
     def isNonEmptyFile(self, field_data, all_data):
-        try:
-            content = field_data['content']
-        except TypeError:
+        if field_data.has_key('_file_upload_error'):
+            raise validators.CriticalValidationError, field_data['_file_upload_error']
+        if not field_data.has_key('filename'):
             raise validators.CriticalValidationError, gettext("No file was submitted. Check the encoding type on the form.")
-        if not content:
+        if not field_data['content-length']:
             raise validators.CriticalValidationError, gettext("The submitted file is empty.")
 
     def render(self, data):
         return '<input type="file" id="%s" class="v%s" name="%s" />' % \
             (self.get_id(), self.__class__.__name__, self.field_name)
 
+    def prepare(self, new_data):
+        if new_data.has_key('_file_upload_error'):
+            # pretend we got something in the field to raise a validation error later
+            new_data[self.field_name] = { '_file_upload_error': new_data['_file_upload_error'] }
+
     def html2python(data):
         if data is None:
             raise EmptyValue

django/db/models/base.py

@@ -12 +12 @@
 from django.dispatch import dispatcher
 from django.utils.datastructures import SortedDict
 from django.utils.functional import curry
+from django.utils.file import file_move_safe
 from django.conf import settings
 from itertools import izip
 import types
 import sys
 import os
 
+                
 class ModelBase(type):
     "Metaclass for all models"
     def __new__(cls, name, bases, attrs):
@@ -361 +363 @@
     def _get_FIELD_size(self, field):
         return os.path.getsize(self._get_FIELD_filename(field))
 
-    def _save_FIELD_file(self, field, filename, raw_contents, save=True):
+    def _save_FIELD_file(self, field, filename, raw_field, save=True):
         directory = field.get_directory_name()
         try: # Create the date-based directory if it doesn't exist.
             os.makedirs(os.path.join(settings.MEDIA_ROOT, directory))
         except OSError: # Directory probably already exists.
             pass
+
+        if filename is None:
+            filename = raw_field['filename']
+
         filename = field.get_filename(filename)
 
         # If the filename already exists, keep adding an underscore to the name of
@@ -383 +389 @@
         setattr(self, field.attname, filename)
 
         full_filename = self._get_FIELD_filename(field)
-        fp = open(full_filename, 'wb')
-        fp.write(raw_contents)
-        fp.close()
+        if raw_field.has_key('tmpfilename'):
+            raw_field['tmpfile'].close()
+            file_move_safe(raw_field['tmpfilename'], full_filename)
+        else:
+            from django.utils import file_locks
+            fp = open(full_filename, 'wb')
+            # exclusive lock
+            file_locks.lock(fp, file_locks.LOCK_EX)
+            fp.write(raw_field['content'])
+            fp.close()
 
         # Save the width and/or height, if applicable.
         if isinstance(field, ImageField) and (field.width_field or field.height_field):

django/db/models/fields/__init__.py

@@ -701 +701 @@
         setattr(cls, 'get_%s_filename' % self.name, curry(cls._get_FIELD_filename, field=self))
         setattr(cls, 'get_%s_url' % self.name, curry(cls._get_FIELD_url, field=self))
         setattr(cls, 'get_%s_size' % self.name, curry(cls._get_FIELD_size, field=self))
-        setattr(cls, 'save_%s_file' % self.name, lambda instance, filename, raw_contents, save=True: instance._save_FIELD_file(self, filename, raw_contents, save))
+        setattr(cls, 'save_%s_file' % self.name, lambda instance, filename, raw_field, save=True: instance._save_FIELD_file(self, filename, raw_field, save))
+        setattr(cls, 'move_%s_file' % self.name, lambda instance, raw_field, save=True: instance._save_FIELD_file(self, None, raw_field, save))        
         dispatcher.connect(self.delete_file, signal=signals.post_delete, sender=cls)
 
     def delete_file(self, instance):
@@ -724 +725 @@
         if new_data.get(upload_field_name, False):
             func = getattr(new_object, 'save_%s_file' % self.name)
             if rel:
-                func(new_data[upload_field_name][0]["filename"], new_data[upload_field_name][0]["content"], save)
+                func(new_data[upload_field_name][0]["filename"], new_data[upload_field_name][0], save)
             else:
-                func(new_data[upload_field_name]["filename"], new_data[upload_field_name]["content"], save)
+                func(new_data[upload_field_name]["filename"], new_data[upload_field_name], save)
 
     def get_directory_name(self):
         return os.path.normpath(datetime.datetime.now().strftime(self.upload_to))

django/conf/global_settings.py

@@ -242 +242 @@
 # isExistingURL validator.
 URL_VALIDATOR_USER_AGENT = "Django/0.96pre (http://www.djangoproject.com)"
 
+# The directory to place streamed file uploads. The web server needs write
+# permissions on this directory.
+# If this is None, streaming uploads are disabled.
+FILE_UPLOAD_DIR = None
+
+# The minimum size of a POST before file uploads are streamed to disk.
+# Any less than this number, and the file is uploaded to memory.
+# Size is in bytes.
+STREAMING_MIN_POST_SIZE = 512 * (2**10)
+
 ##############
 # MIDDLEWARE #
 ##############

django/core/handlers/wsgi.py

@@ -75 +75 @@
         self.environ = environ
         self.path = environ['PATH_INFO']
         self.META = environ
+        self.META['UPLOAD_PROGRESS_ID'] = self._get_file_progress_id()
         self.method = environ['REQUEST_METHOD'].upper()
 
     def __repr__(self):
@@ -111 +112 @@
             if self.environ.get('CONTENT_TYPE', '').startswith('multipart'):
                 header_dict = dict([(k, v) for k, v in self.environ.items() if k.startswith('HTTP_')])
                 header_dict['Content-Type'] = self.environ.get('CONTENT_TYPE', '')
-                self._post, self._files = http.parse_file_upload(header_dict, self.raw_post_data)
+                header_dict['Content-Length'] = self.environ.get('CONTENT_LENGTH', '')
+                header_dict['X-Progress-ID'] = self.environ.get('HTTP_X_PROGRESS_ID', '')
+                try:
+                    self._post, self._files = http.parse_file_upload(header_dict, self.environ['wsgi.input'], self)
+                except:
+                    self._post, self._files = {}, {} # make sure we dont read the input stream again
+                    raise
+                self._raw_post_data = None # raw data is not available for streamed multipart messages
             else:
                 self._post, self._files = http.QueryDict(self.raw_post_data), datastructures.MultiValueDict()
         else:
@@ -167 +175 @@
             buf.close()
             return self._raw_post_data
 
+    def _get_file_progress_id(self):
+        """
+        Returns the Progress ID of the request,
+        usually provided if there is a file upload
+        going on.
+        Returns ``None`` if no progress ID is specified.
+        """
+        return self._get_file_progress_from_args(self.environ,
+                                                 self.GET,
+                                                 self.environ.get('QUERY_STRING', ''))
+
     GET = property(_get_get, _set_get)
     POST = property(_get_post, _set_post)
     COOKIES = property(_get_cookies, _set_cookies)

django/core/handlers/base.py

@@ -1 +1 @@
 from django.core import signals
 from django.dispatch import dispatcher
 from django import http
-import sys
+import sys, re
 
 class BaseHandler(object):
     def __init__(self):
@@ -129 +129 @@
         "Helper function to return the traceback as a string"
         import traceback
         return '\n'.join(traceback.format_exception(*(exc_info or sys.exc_info())))
+
+

django/core/handlers/modpython.py

@@ -47 +47 @@
     def _load_post_and_files(self):
         "Populates self._post and self._files"
         if 'content-type' in self._req.headers_in and self._req.headers_in['content-type'].startswith('multipart'):
-            self._post, self._files = http.parse_file_upload(self._req.headers_in, self.raw_post_data)
+            self._raw_post_data = None # raw data is not available for streamed multipart messages
+            try:
+                self._post, self._files = http.parse_file_upload(self._req.headers_in, self._req, self)
+            except:
+                self._post, self._files = {}, {} # make sure we dont read the input stream again
+                raise
         else:
             self._post, self._files = http.QueryDict(self.raw_post_data), datastructures.MultiValueDict()
 
@@ -92 +97 @@
                 'AUTH_TYPE':         self._req.ap_auth_type,
                 'CONTENT_LENGTH':    self._req.clength, # This may be wrong
                 'CONTENT_TYPE':      self._req.content_type, # This may be wrong
-                'GATEWAY_INTERFACE': 'CGI/1.1',
-                'PATH_INFO':         self._req.path_info,
-                'PATH_TRANSLATED':   None, # Not supported
-                'QUERY_STRING':      self._req.args,
-                'REMOTE_ADDR':       self._req.connection.remote_ip,
-                'REMOTE_HOST':       None, # DNS lookups not supported
-                'REMOTE_IDENT':      self._req.connection.remote_logname,
-                'REMOTE_USER':       self._req.user,
-                'REQUEST_METHOD':    self._req.method,
-                'SCRIPT_NAME':       None, # Not supported
-                'SERVER_NAME':       self._req.server.server_hostname,
-                'SERVER_PORT':       self._req.server.port,
-                'SERVER_PROTOCOL':   self._req.protocol,
-                'SERVER_SOFTWARE':   'mod_python'
+                'GATEWAY_INTERFACE':  'CGI/1.1',
+                'PATH_INFO':          self._req.path_info,
+                'PATH_TRANSLATED':    None, # Not supported
+                'QUERY_STRING':       self._req.args,
+                'REMOTE_ADDR':        self._req.connection.remote_ip,
+                'REMOTE_HOST':        None, # DNS lookups not supported
+                'REMOTE_IDENT':       self._req.connection.remote_logname,
+                'REMOTE_USER':        self._req.user,
+                'REQUEST_METHOD':     self._req.method,
+                'SCRIPT_NAME':        None, # Not supported
+                'SERVER_NAME':        self._req.server.server_hostname,
+                'SERVER_PORT':        self._req.server.port,
+                'SERVER_PROTOCOL':    self._req.protocol,
+                'UPLOAD_PROGRESS_ID': self._get_file_progress_id(),
+                'SERVER_SOFTWARE':    'mod_python'
             }
             for key, value in self._req.headers_in.items():
                 key = 'HTTP_' + key.upper().replace('-', '_')
@@ -122 +128 @@
     def _get_method(self):
         return self.META['REQUEST_METHOD'].upper()
 
+    def _get_file_progress_id(self):
+        """
+        Returns the Progress ID of the request,
+        usually provided if there is a file upload
+        going on.
+        Returns ``None`` if no progress ID is specified.
+        """
+        return self._get_file_progress_from_args(self._req.headers_in,
+                                                 self.GET,
+                                                 self._req.args)
+
     GET = property(_get_get, _set_get)
     POST = property(_get_post, _set_post)
     COOKIES = property(_get_cookies, _set_cookies)

django/utils/file_locks.py

@@ +1 @@
+"""
+Locking portability by Jonathan Feignberg <jdf@pobox.com> in python cookbook
+
+Example Usage::
+
+    from django.utils import file_locks
+
+    f = open('./file', 'wb')
+
+    file_locks.lock(f, file_locks.LOCK_EX)
+    f.write('Django')
+    f.close()
+"""
+
+
+import os
+
+__all__ = ['LOCK_EX','LOCK_SH','LOCK_NB','lock','unlock']
+
+if os.name == 'nt':
+        import win32con
+        import win32file
+        import pywintypes
+        LOCK_EX = win32con.LOCKFILE_EXCLUSIVE_LOCK
+        LOCK_SH = 0
+        LOCK_NB = win32con.LOCKFILE_FAIL_IMMEDIATELY
+        __overlapped = pywintypes.OVERLAPPED()
+elif os.name == 'posix':
+        import fcntl
+        LOCK_EX = fcntl.LOCK_EX
+        LOCK_SH = fcntl.LOCK_SH
+        LOCK_NB = fcntl.LOCK_NB
+else:
+        raise RuntimeError("Locking only defined for nt and posix platforms")
+
+if os.name == 'nt':
+        def lock(file, flags):
+                hfile = win32file._get_osfhandle(file.fileno())
+                win32file.LockFileEx(hfile, flags, 0, -0x10000, __overlapped)
+
+        def unlock(file):
+                hfile = win32file._get_osfhandle(file.fileno())
+                win32file.UnlockFileEx(hfile, 0, -0x10000, __overlapped)
+
+elif os.name =='posix':
+        def lock(file, flags):
+                fcntl.flock(file.fileno(), flags)
+
+        def unlock(file):
+                fcntl.flock(file.fileno(), fcntl.LOCK_UN)

django/utils/file.py

@@ +1 @@
+import os
+
+__all__ = ['file_move_safe']
+
+try:
+    import shutil
+    file_move = shutil.move
+except ImportError:
+    file_move = os.rename
+
+def file_move_safe(old_file_name, new_file_name, chunk_size = 1024*64, allow_overwrite=False):
+    """
+    Moves a file from one location to another in the safest way possible.
+    
+    First, it tries using shutils.move, which is OS-dependent but doesn't
+    break with change of filesystems. Then it tries os.rename, which will
+    break if it encounters a change in filesystems. Lastly, it streams
+    it manually from one file to another in python.
+
+    Without ``allow_overwrite``, if the destination file exists, the
+    file will raise an IOError.
+    """
+
+    from django.utils import file_locks
+
+    if old_file_name == new_file_name:
+        # No file moving takes place.
+        return
+
+    if not allow_overwrite and os.path.exists(new_file_name):
+        raise IOError, "Django does not allow overwriting files."
+
+    try:
+        file_move(old_file_name, new_file_name)
+        return
+    except OSError: # moving to another filesystem
+        pass
+
+    new_file = open(new_file_name, 'wb')
+    # exclusive lock
+    file_locks.lock(new_file, file_locks.LOCK_EX)
+    old_file = open(old_file_name, 'rb')
+    current_chunk = None
+
+    while current_chunk != '':
+        current_chunk = old_file.read(chunk_size)
+        new_file.write(current_chunk)
+
+    new_file.close()
+    old_file.close()
+
+    os.remove(old_file_name)
+

tests/modeltests/test_client/views.py

@@ -46 +46 @@
 
     return HttpResponse(t.render(c))
 
+def post_file_view(request):
+    "A view that expects a multipart post and returns a file in the context"
+    t = Template('File {{ file.filename }} received', name='POST Template')
+    c = Context({'file': request.FILES['file_file']})
+    return HttpResponse(t.render(c))
+
 def redirect_view(request):
     "A view that redirects all requests to the GET view"
     return HttpResponseRedirect('/test_client/get_view/')

tests/modeltests/test_client/models.py

@@ -3 +3 @@
 
 The test client is a class that can act like a simple
 browser for testing purposes.
-  
+
 It allows the user to compose GET and POST requests, and
 obtain the response that the server gave to those requests.
 The server Response objects are annotated with the details
@@ -76 +76 @@
         self.assertEqual(response.template.name, "Book template")
         self.assertEqual(response.content, "Blink - Malcolm Gladwell")
 
+    def test_post_file_view(self):
+        "POST this python file to a view"
+        import os, tempfile
+        from django.conf import settings
+        file = __file__.replace('.pyc', '.py')
+        for upload_dir in [None, tempfile.gettempdir()]:
+            settings.FILE_UPLOAD_DIR = upload_dir
+            post_data = { 'name': file, 'file': open(file) }
+            response = self.client.post('/test_client/post_file_view/', post_data)
+            self.failUnless('models.py' in response.context['file']['filename'])
+            self.failUnless(len(response.context['file']['content']) == os.path.getsize(file))
+            if upload_dir:
+                self.failUnless(response.context['file']['tmpfilename'])
+
     def test_redirect(self):
         "GET a URL that redirects elsewhere"
         response = self.client.get('/test_client/redirect_view/')

tests/modeltests/test_client/urls.py

@@ -5 +5 @@
 urlpatterns = patterns('',
     (r'^get_view/$', views.get_view),
     (r'^post_view/$', views.post_view),
+    (r'^post_file_view/$', views.post_file_view),
     (r'^raw_post_view/$', views.raw_post_view),
     (r'^redirect_view/$', views.redirect_view),
     (r'^permanent_redirect_view/$', redirect_to, { 'url': '/test_client/get_view/' }),

docs/request_response.txt

@@ -72 +72 @@
 ``FILES``
     A dictionary-like object containing all uploaded files. Each key in
     ``FILES`` is the ``name`` from the ``<input type="file" name="" />``. Each
-    value in ``FILES`` is a standard Python dictionary with the following three
+    value in ``FILES`` is a standard Python dictionary with the following four
     keys:
 
         * ``filename`` -- The name of the uploaded file, as a Python string.
         * ``content-type`` -- The content type of the uploaded file.
         * ``content`` -- The raw content of the uploaded file.
+        * ``content-length`` -- The length of the content in bytes.
 
+    If streaming file uploads are enabled two additional keys
+    describing the uploaded file will be present:
+
+        * ``tmpfilename`` -- The filename for the temporary file.
+        * ``tmpfile`` -- An open file object for the temporary file.
+
+    The temporary file will be removed when the request finishes.
+
+    Note that accessing ``content`` when streaming uploads are enabled
+    will read the whole file into memory which may not be what you want.
+
     Note that ``FILES`` will only contain data if the request method was POST
     and the ``<form>`` that posted to the request had
     ``enctype="multipart/form-data"``. Otherwise, ``FILES`` will be a blank

docs/settings.txt

@@ -448 +448 @@
 
 .. _Testing Django Applications: ../testing/
 
+FILE_UPLOAD_DIR
+---------------
+
+Default: ``None``
+
+Path to a directory where temporary files should be written during
+file uploads. Leaving this as ``None`` will disable streaming file uploads,
+and cause all uploaded files to be stored (temporarily) in memory.
+
 IGNORABLE_404_ENDS
 ------------------
 
@@ -764 +773 @@
 
 .. _site framework docs: ../sites/
 
+STREAMING_MIN_POST_SIZE
+-----------------------
+
+Default: 524288 (``512*1024``)
+
+An integer specifying the minimum number of bytes that has to be
+received (in a POST) for file upload streaming to take place. Any
+request smaller than this will be handled in memory. 
+Note: ``FILE_UPLOAD_DIR`` has to be defined to enable streaming.
+
 TEMPLATE_CONTEXT_PROCESSORS
 ---------------------------
 

docs/forms.txt

@@ -475 +475 @@
    new_data = request.POST.copy()
    new_data.update(request.FILES)
 
+Streaming file uploads.
+-----------------------
+
+File uploads will be read into memory by default. This works fine for
+small to medium sized uploads (from 1MB to 100MB depending on your
+setup and usage). If you want to support larger uploads you can enable
+upload streaming where only a small part of the file will be in memory
+at any time. To do this you need to specify the ``FILE_UPLOAD_DIR``
+setting (see the settings_ document for more details).
+
+See `request object`_ for more details about ``request.FILES`` objects
+with streaming file uploads enabled.
+
 Validators
 ==========
 
@@ -698 +711 @@
 .. _`generic views`: ../generic_views/
 .. _`models API`: ../model-api/
 .. _settings: ../settings/
+.. _request object: ../request_response/#httprequest-objects