# Copyright (c) 2010-2014 OpenStack Foundation. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or # implied. # See the License for the specific language governing permissions and # limitations under the License. """ Implementation of S3 Multipart Upload. This module implements S3 Multipart Upload APIs with the Swift SLO feature. The following explains how S3api uses swift container and objects to store S3 upload information: ----------------- [bucket]+segments ----------------- A container to store upload information. [bucket] is the original bucket where multipart upload is initiated. ----------------------------- [bucket]+segments/[upload_id] ----------------------------- An object of the ongoing upload id. The object is empty and used for checking the target upload status. If the object exists, it means that the upload is initiated but not either completed or aborted. ------------------------------------------- [bucket]+segments/[upload_id]/[part_number] ------------------------------------------- The last suffix is the part number under the upload id. When the client uploads the parts, they will be stored in the namespace with [bucket]+segments/[upload_id]/[part_number]. Example listing result in the [bucket]+segments container:: [bucket]+segments/[upload_id1] # upload id object for upload_id1 [bucket]+segments/[upload_id1]/1 # part object for upload_id1 [bucket]+segments/[upload_id1]/2 # part object for upload_id1 [bucket]+segments/[upload_id1]/3 # part object for upload_id1 [bucket]+segments/[upload_id2] # upload id object for upload_id2 [bucket]+segments/[upload_id2]/1 # part object for upload_id2 [bucket]+segments/[upload_id2]/2 # part object for upload_id2 . . Those part objects are directly used as segments of a Swift Static Large Object when the multipart upload is completed. """ from hashlib import md5 import os import re from swift.common.swob import Range from swift.common.utils import json, public, reiterate from swift.common.db import utf8encode from six.moves.urllib.parse import quote, urlparse from swift.common.middleware.s3api.controllers.base import Controller, \ bucket_operation, object_operation, check_container_existence from swift.common.middleware.s3api.s3response import InvalidArgument, \ ErrorResponse, MalformedXML, \ InvalidPart, BucketAlreadyExists, EntityTooSmall, InvalidPartOrder, \ InvalidRequest, HTTPOk, HTTPNoContent, NoSuchKey, NoSuchUpload, \ NoSuchBucket, BucketAlreadyOwnedByYou from swift.common.middleware.s3api.exception import BadSwiftRequest from swift.common.middleware.s3api.utils import unique_id, \ MULTIUPLOAD_SUFFIX, S3Timestamp, sysmeta_header from swift.common.middleware.s3api.etree import Element, SubElement, \ fromstring, tostring, XMLSyntaxError, DocumentInvalid DEFAULT_MAX_PARTS_LISTING = 1000 DEFAULT_MAX_UPLOADS = 1000 MAX_COMPLETE_UPLOAD_BODY_SIZE = 2048 * 1024 def _get_upload_info(req, app, upload_id): container = req.container_name + MULTIUPLOAD_SUFFIX obj = '%s/%s' % (req.object_name, upload_id) try: return req.get_response(app, 'HEAD', container=container, obj=obj) except NoSuchKey: raise NoSuchUpload(upload_id=upload_id) def _check_upload_info(req, app, upload_id): _get_upload_info(req, app, upload_id) class PartController(Controller): """ Handles the following APIs: * Upload Part * Upload Part - Copy Those APIs are logged as PART operations in the S3 server log. """ @public @object_operation @check_container_existence def PUT(self, req): """ Handles Upload Part and Upload Part Copy. """ if 'uploadId' not in req.params: raise InvalidArgument('ResourceType', 'partNumber', 'Unexpected query string parameter') try: part_number = int(req.params['partNumber']) if part_number < 1 or self.conf.max_upload_part_num < part_number: raise Exception() except Exception: err_msg = 'Part number must be an integer between 1 and %d,' \ ' inclusive' % self.conf.max_upload_part_num raise InvalidArgument('partNumber', req.params['partNumber'], err_msg) upload_id = req.params['uploadId'] _check_upload_info(req, self.app, upload_id) req.container_name += MULTIUPLOAD_SUFFIX req.object_name = '%s/%s/%d' % (req.object_name, upload_id, part_number) req_timestamp = S3Timestamp.now() req.headers['X-Timestamp'] = req_timestamp.internal source_resp = req.check_copy_source(self.app) if 'X-Amz-Copy-Source' in req.headers and \ 'X-Amz-Copy-Source-Range' in req.headers: rng = req.headers['X-Amz-Copy-Source-Range'] header_valid = True try: rng_obj = Range(rng) if len(rng_obj.ranges) != 1: header_valid = False except ValueError: header_valid = False if not header_valid: err_msg = ('The x-amz-copy-source-range value must be of the ' 'form bytes=first-last where first and last are ' 'the zero-based offsets of the first and last ' 'bytes to copy') raise InvalidArgument('x-amz-source-range', rng, err_msg) source_size = int(source_resp.headers['Content-Length']) if not rng_obj.ranges_for_length(source_size): err_msg = ('Range specified is not valid for source object ' 'of size: %s' % source_size) raise InvalidArgument('x-amz-source-range', rng, err_msg) req.headers['Range'] = rng del req.headers['X-Amz-Copy-Source-Range'] resp = req.get_response(self.app) if 'X-Amz-Copy-Source' in req.headers: resp.append_copy_resp_body(req.controller_name, req_timestamp.s3xmlformat) resp.status = 200 return resp class UploadsController(Controller): """ Handles the following APIs: * List Multipart Uploads * Initiate Multipart Upload Those APIs are logged as UPLOADS operations in the S3 server log. """ @public @bucket_operation(err_resp=InvalidRequest, err_msg="Key is not expected for the GET method " "?uploads subresource") @check_container_existence def GET(self, req): """ Handles List Multipart Uploads """ def separate_uploads(uploads, prefix, delimiter): """ separate_uploads will separate uploads into non_delimited_uploads (a subset of uploads) and common_prefixes according to the specified delimiter. non_delimited_uploads is a list of uploads which exclude the delimiter. common_prefixes is a set of prefixes prior to the specified delimiter. Note that the prefix in the common_prefixes includes the delimiter itself. i.e. if '/' delimiter specified and then the uploads is consists of ['foo', 'foo/bar'], this function will return (['foo'], ['foo/']). :param uploads: A list of uploads dictionary :param prefix: A string of prefix reserved on the upload path. (i.e. the delimiter must be searched behind the prefix) :param delimiter: A string of delimiter to split the path in each upload :return (non_delimited_uploads, common_prefixes) """ (prefix, delimiter) = \ utf8encode(prefix, delimiter) non_delimited_uploads = [] common_prefixes = set() for upload in uploads: key = upload['key'] end = key.find(delimiter, len(prefix)) if end >= 0: common_prefix = key[:end + len(delimiter)] common_prefixes.add(common_prefix) else: non_delimited_uploads.append(upload) return non_delimited_uploads, sorted(common_prefixes) encoding_type = req.params.get('encoding-type') if encoding_type is not None and encoding_type != 'url': err_msg = 'Invalid Encoding Method specified in Request' raise InvalidArgument('encoding-type', encoding_type, err_msg) keymarker = req.params.get('key-marker', '') uploadid = req.params.get('upload-id-marker', '') maxuploads = req.get_validated_param( 'max-uploads', DEFAULT_MAX_UPLOADS, DEFAULT_MAX_UPLOADS) query = { 'format': 'json', 'limit': maxuploads + 1, } if uploadid and keymarker: query.update({'marker': '%s/%s' % (keymarker, uploadid)}) elif keymarker: query.update({'marker': '%s/~' % (keymarker)}) if 'prefix' in req.params: query.update({'prefix': req.params['prefix']}) container = req.container_name + MULTIUPLOAD_SUFFIX try: resp = req.get_response(self.app, container=container, query=query) objects = json.loads(resp.body) except NoSuchBucket: # Assume NoSuchBucket as no uploads objects = [] def object_to_upload(object_info): obj, upid = object_info['name'].rsplit('/', 1) obj_dict = {'key': obj, 'upload_id': upid, 'last_modified': object_info['last_modified']} return obj_dict # uploads is a list consists of dict, {key, upload_id, last_modified} # Note that pattern matcher will drop whole segments objects like as # object_name/upload_id/1. pattern = re.compile('/[0-9]+$') uploads = [object_to_upload(obj) for obj in objects if pattern.search(obj.get('name', '')) is None] prefixes = [] if 'delimiter' in req.params: prefix = req.params.get('prefix', '') delimiter = req.params['delimiter'] uploads, prefixes = \ separate_uploads(uploads, prefix, delimiter) if len(uploads) > maxuploads: uploads = uploads[:maxuploads] truncated = True else: truncated = False nextkeymarker = '' nextuploadmarker = '' if len(uploads) > 1: nextuploadmarker = uploads[-1]['upload_id'] nextkeymarker = uploads[-1]['key'] result_elem = Element('ListMultipartUploadsResult') SubElement(result_elem, 'Bucket').text = req.container_name SubElement(result_elem, 'KeyMarker').text = keymarker SubElement(result_elem, 'UploadIdMarker').text = uploadid SubElement(result_elem, 'NextKeyMarker').text = nextkeymarker SubElement(result_elem, 'NextUploadIdMarker').text = nextuploadmarker if 'delimiter' in req.params: SubElement(result_elem, 'Delimiter').text = \ req.params['delimiter'] if 'prefix' in req.params: SubElement(result_elem, 'Prefix').text = req.params['prefix'] SubElement(result_elem, 'MaxUploads').text = str(maxuploads) if encoding_type is not None: SubElement(result_elem, 'EncodingType').text = encoding_type SubElement(result_elem, 'IsTruncated').text = \ 'true' if truncated else 'false' # TODO: don't show uploads which are initiated before this bucket is # created. for u in uploads: upload_elem = SubElement(result_elem, 'Upload') name = u['key'] if encoding_type == 'url': name = quote(name) SubElement(upload_elem, 'Key').text = name SubElement(upload_elem, 'UploadId').text = u['upload_id'] initiator_elem = SubElement(upload_elem, 'Initiator') SubElement(initiator_elem, 'ID').text = req.user_id SubElement(initiator_elem, 'DisplayName').text = req.user_id owner_elem = SubElement(upload_elem, 'Owner') SubElement(owner_elem, 'ID').text = req.user_id SubElement(owner_elem, 'DisplayName').text = req.user_id SubElement(upload_elem, 'StorageClass').text = 'STANDARD' SubElement(upload_elem, 'Initiated').text = \ u['last_modified'][:-3] + 'Z' for p in prefixes: elem = SubElement(result_elem, 'CommonPrefixes') SubElement(elem, 'Prefix').text = p body = tostring(result_elem) return HTTPOk(body=body, content_type='application/xml') @public @object_operation @check_container_existence def POST(self, req): """ Handles Initiate Multipart Upload. """ # Create a unique S3 upload id from UUID to avoid duplicates. upload_id = unique_id() container = req.container_name + MULTIUPLOAD_SUFFIX content_type = req.headers.get('Content-Type') if content_type: req.headers[sysmeta_header('object', 'has-content-type')] = 'yes' req.headers[ sysmeta_header('object', 'content-type')] = content_type else: req.headers[sysmeta_header('object', 'has-content-type')] = 'no' req.headers['Content-Type'] = 'application/directory' try: req.get_response(self.app, 'PUT', container, '') except (BucketAlreadyExists, BucketAlreadyOwnedByYou): pass obj = '%s/%s' % (req.object_name, upload_id) req.headers.pop('Etag', None) req.headers.pop('Content-Md5', None) req.get_response(self.app, 'PUT', container, obj, body='') result_elem = Element('InitiateMultipartUploadResult') SubElement(result_elem, 'Bucket').text = req.container_name SubElement(result_elem, 'Key').text = req.object_name SubElement(result_elem, 'UploadId').text = upload_id body = tostring(result_elem) return HTTPOk(body=body, content_type='application/xml') class UploadController(Controller): """ Handles the following APIs: * List Parts * Abort Multipart Upload * Complete Multipart Upload Those APIs are logged as UPLOAD operations in the S3 server log. """ @public @object_operation @check_container_existence def GET(self, req): """ Handles List Parts. """ def filter_part_num_marker(o): try: num = int(os.path.basename(o['name'])) return num > part_num_marker except ValueError: return False encoding_type = req.params.get('encoding-type') if encoding_type is not None and encoding_type != 'url': err_msg = 'Invalid Encoding Method specified in Request' raise InvalidArgument('encoding-type', encoding_type, err_msg) upload_id = req.params['uploadId'] _check_upload_info(req, self.app, upload_id) maxparts = req.get_validated_param( 'max-parts', DEFAULT_MAX_PARTS_LISTING, self.conf.max_parts_listing) part_num_marker = req.get_validated_param( 'part-number-marker', 0) query = { 'format': 'json', 'limit': maxparts + 1, 'prefix': '%s/%s/' % (req.object_name, upload_id), 'delimiter': '/' } container = req.container_name + MULTIUPLOAD_SUFFIX resp = req.get_response(self.app, container=container, obj='', query=query) objects = json.loads(resp.body) last_part = 0 # If the caller requested a list starting at a specific part number, # construct a sub-set of the object list. objList = filter(filter_part_num_marker, objects) # pylint: disable-msg=E1103 objList.sort(key=lambda o: int(o['name'].split('/')[-1])) if len(objList) > maxparts: objList = objList[:maxparts] truncated = True else: truncated = False # TODO: We have to retrieve object list again when truncated is True # and some objects filtered by invalid name because there could be no # enough objects for limit defined by maxparts. if objList: o = objList[-1] last_part = os.path.basename(o['name']) result_elem = Element('ListPartsResult') SubElement(result_elem, 'Bucket').text = req.container_name name = req.object_name if encoding_type == 'url': name = quote(name) SubElement(result_elem, 'Key').text = name SubElement(result_elem, 'UploadId').text = upload_id initiator_elem = SubElement(result_elem, 'Initiator') SubElement(initiator_elem, 'ID').text = req.user_id SubElement(initiator_elem, 'DisplayName').text = req.user_id owner_elem = SubElement(result_elem, 'Owner') SubElement(owner_elem, 'ID').text = req.user_id SubElement(owner_elem, 'DisplayName').text = req.user_id SubElement(result_elem, 'StorageClass').text = 'STANDARD' SubElement(result_elem, 'PartNumberMarker').text = str(part_num_marker) SubElement(result_elem, 'NextPartNumberMarker').text = str(last_part) SubElement(result_elem, 'MaxParts').text = str(maxparts) if 'encoding-type' in req.params: SubElement(result_elem, 'EncodingType').text = \ req.params['encoding-type'] SubElement(result_elem, 'IsTruncated').text = \ 'true' if truncated else 'false' for i in objList: part_elem = SubElement(result_elem, 'Part') SubElement(part_elem, 'PartNumber').text = i['name'].split('/')[-1] SubElement(part_elem, 'LastModified').text = \ i['last_modified'][:-3] + 'Z' SubElement(part_elem, 'ETag').text = '"%s"' % i['hash'] SubElement(part_elem, 'Size').text = str(i['bytes']) body = tostring(result_elem) return HTTPOk(body=body, content_type='application/xml') @public @object_operation @check_container_existence def DELETE(self, req): """ Handles Abort Multipart Upload. """ upload_id = req.params['uploadId'] _check_upload_info(req, self.app, upload_id) # First check to see if this multi-part upload was already # completed. Look in the primary container, if the object exists, # then it was completed and we return an error here. container = req.container_name + MULTIUPLOAD_SUFFIX obj = '%s/%s' % (req.object_name, upload_id) req.get_response(self.app, container=container, obj=obj) # The completed object was not found so this # must be a multipart upload abort. # We must delete any uploaded segments for this UploadID and then # delete the object in the main container as well query = { 'format': 'json', 'prefix': '%s/%s/' % (req.object_name, upload_id), 'delimiter': '/', } resp = req.get_response(self.app, 'GET', container, '', query=query) # Iterate over the segment objects and delete them individually objects = json.loads(resp.body) for o in objects: container = req.container_name + MULTIUPLOAD_SUFFIX req.get_response(self.app, container=container, obj=o['name']) return HTTPNoContent() @public @object_operation @check_container_existence def POST(self, req): """ Handles Complete Multipart Upload. """ upload_id = req.params['uploadId'] resp = _get_upload_info(req, self.app, upload_id) headers = {'Accept': 'application/json'} for key, val in resp.headers.items(): _key = key.lower() if _key.startswith('x-amz-meta-'): headers['x-object-meta-' + _key[11:]] = val hct_header = sysmeta_header('object', 'has-content-type') if resp.sysmeta_headers.get(hct_header) == 'yes': content_type = resp.sysmeta_headers.get( sysmeta_header('object', 'content-type')) elif hct_header in resp.sysmeta_headers: # has-content-type is present but false, so no content type was # set on initial upload. In that case, we won't set one on our # PUT request. Swift will end up guessing one based on the # object name. content_type = None else: content_type = resp.headers.get('Content-Type') if content_type: headers['Content-Type'] = content_type # Query for the objects in the segments area to make sure it completed query = { 'format': 'json', 'prefix': '%s/%s/' % (req.object_name, upload_id), 'delimiter': '/' } container = req.container_name + MULTIUPLOAD_SUFFIX resp = req.get_response(self.app, 'GET', container, '', query=query) objinfo = json.loads(resp.body) objtable = dict((o['name'], {'path': '/'.join(['', container, o['name']]), 'etag': o['hash'], 'size_bytes': o['bytes']}) for o in objinfo) s3_etag_hasher = md5() manifest = [] previous_number = 0 try: xml = req.xml(MAX_COMPLETE_UPLOAD_BODY_SIZE) if not xml: raise InvalidRequest(msg='You must specify at least one part') complete_elem = fromstring( xml, 'CompleteMultipartUpload', self.logger) for part_elem in complete_elem.iterchildren('Part'): part_number = int(part_elem.find('./PartNumber').text) if part_number <= previous_number: raise InvalidPartOrder(upload_id=upload_id) previous_number = part_number etag = part_elem.find('./ETag').text if len(etag) >= 2 and etag[0] == '"' and etag[-1] == '"': # strip double quotes etag = etag[1:-1] info = objtable.get("%s/%s/%s" % (req.object_name, upload_id, part_number)) if info is None or info['etag'] != etag: raise InvalidPart(upload_id=upload_id, part_number=part_number) s3_etag_hasher.update(etag.decode('hex')) info['size_bytes'] = int(info['size_bytes']) manifest.append(info) except (XMLSyntaxError, DocumentInvalid): # NB: our schema definitions catch uploads with no parts here raise MalformedXML() except ErrorResponse: raise except Exception as e: self.logger.error(e) raise s3_etag = '%s-%d' % (s3_etag_hasher.hexdigest(), len(manifest)) headers[sysmeta_header('object', 'etag')] = s3_etag # Leave base header value blank; SLO will populate c_etag = '; s3_etag=%s' % s3_etag headers['X-Object-Sysmeta-Container-Update-Override-Etag'] = c_etag # Check the size of each segment except the last and make sure they are # all more than the minimum upload chunk size for info in manifest[:-1]: if info['size_bytes'] < self.conf.min_segment_size: raise EntityTooSmall() def response_iter(): # NB: XML requires that the XML declaration, if present, be at the # very start of the document. Clients *will* call us out on not # being valid XML if we pass through whitespace before it. # Track whether we've sent anything yet so we can yield out that # declaration *first* yielded_anything = False try: try: # TODO: add support for versioning put_resp = req.get_response( self.app, 'PUT', body=json.dumps(manifest), query={'multipart-manifest': 'put', 'heartbeat': 'on'}, headers=headers) if put_resp.status_int == 202: body = [] put_resp.fix_conditional_response() for chunk in put_resp.response_iter: if not chunk.strip(): if not yielded_anything: yield ('\n') yielded_anything = True yield chunk body.append(chunk) body = json.loads(''.join(body)) if body['Response Status'] != '201 Created': raise InvalidRequest( status=body['Response Status'], msg='\n'.join(': '.join(err) for err in body['Errors'])) except BadSwiftRequest as e: msg = str(e) expected_msg = ('too small; each segment must be ' 'at least 1 byte') if expected_msg in msg: # FIXME: AWS S3 allows a smaller object than 5 MB if # there is only one part. Use a COPY request to copy # the part object from the segments container instead. raise EntityTooSmall(msg) else: raise # clean up the multipart-upload record obj = '%s/%s' % (req.object_name, upload_id) try: req.get_response(self.app, 'DELETE', container, obj) except NoSuchKey: # We know that this existed long enough for us to HEAD pass result_elem = Element('CompleteMultipartUploadResult') # NOTE: boto with sig v4 appends port to HTTP_HOST value at # the request header when the port is non default value and it # makes req.host_url like as http://localhost:8080:8080/path # that obviously invalid. Probably it should be resolved at # swift.common.swob though, tentatively we are parsing and # reconstructing the correct host_url info here. # in detail, https://github.com/boto/boto/pull/3513 parsed_url = urlparse(req.host_url) host_url = '%s://%s' % (parsed_url.scheme, parsed_url.hostname) if parsed_url.port: host_url += ':%s' % parsed_url.port SubElement(result_elem, 'Location').text = host_url + req.path SubElement(result_elem, 'Bucket').text = req.container_name SubElement(result_elem, 'Key').text = req.object_name SubElement(result_elem, 'ETag').text = '"%s"' % s3_etag resp.headers.pop('ETag', None) if yielded_anything: yield '\n' yield tostring(result_elem, xml_declaration=not yielded_anything) except ErrorResponse as err_resp: if yielded_anything: err_resp.xml_declaration = False yield '\n' else: # Oh good, we can still change HTTP status code, too! resp.status = err_resp.status for chunk in err_resp({}, lambda *a: None): yield chunk resp = HTTPOk() # assume we're good for now... but see above! resp.app_iter = reiterate(response_iter()) resp.content_type = "application/xml" return resp