def _fetch_sub_slo_segments(self, req, version, acc, con, obj):
"""
Fetch the submanifest, parse it, and return it.
Raise exception on failures.
"""
sub_req = make_request(
req.environ, path='/'.join(['', version, acc, con, obj]),
method='GET',
headers={'x-auth-token': req.headers.get('x-auth-token')},
agent=('%(orig)s ' + 'SLO MultipartGET'), swift_source='SLO')
sub_resp = sub_req.get_response(self.slo.app)
if not is_success(sub_resp.status_int):
raise ListingIterError(
'ERROR: while fetching %s, GET of submanifest %s '
'failed with status %d' % (req.path, sub_req.path,
sub_resp.status_int))
try:
with closing_if_possible(sub_resp.app_iter):
return json.loads(''.join(sub_resp.app_iter))
except ValueError as err:
raise ListingIterError(
'ERROR: while fetching %s, JSON-decoding of submanifest %s '
'failed with %s' % (req.path, sub_req.path, err))
def _fetch_sub_slo_segments(self, req, version, acc, con, obj):
"""
Fetch the submanifest, parse it, and return it.
Raise exception on failures.
"""
sub_req = make_request(
req.environ,
path='/'.join(['', version, acc, con, obj]),
method='GET',
headers={'x-auth-token': req.headers.get('x-auth-token')},
agent=('%(orig)s ' + 'SLO MultipartGET'),
swift_source='SLO')
sub_resp = sub_req.get_response(self.slo.app)
if not is_success(sub_resp.status_int):
raise ListingIterError(
'ERROR: while fetching %s, GET of submanifest %s '
'failed with status %d' %
(req.path, sub_req.path, sub_resp.status_int))
try:
with closing_if_possible(sub_resp.app_iter):
return json.loads(''.join(sub_resp.app_iter))
except ValueError as err:
raise ListingIterError(
'ERROR: while fetching %s, JSON-decoding of submanifest %s '
'failed with %s' % (req.path, sub_req.path, err))
def handle_slo_get_or_head(self, req, start_response):
"""
Takes a request and a start_response callable and does the normal WSGI
thing with them. Returns an iterator suitable for sending up the WSGI
chain.
:param req: swob.Request object; is a GET or HEAD request aimed at
what may be a static large object manifest (or may not).
:param start_response: WSGI start_response callable
"""
resp_iter = self._app_call(req.environ)
# make sure this response is for a static large object manifest
for header, value in self._response_headers:
if (header.lower() == 'x-static-large-object' and
config_true_value(value)):
break
else:
# Not a static large object manifest. Just pass it through.
start_response(self._response_status,
self._response_headers,
self._response_exc_info)
return resp_iter
# Handle pass-through request for the manifest itself
if req.params.get('multipart-manifest') == 'get':
new_headers = []
for header, value in self._response_headers:
if header.lower() == 'content-type':
new_headers.append(('Content-Type',
'application/json; charset=utf-8'))
else:
new_headers.append((header, value))
self._response_headers = new_headers
start_response(self._response_status,
self._response_headers,
self._response_exc_info)
return resp_iter
if self._need_to_refetch_manifest(req):
req.environ['swift.non_client_disconnect'] = True
close_if_possible(resp_iter)
del req.environ['swift.non_client_disconnect']
get_req = make_request(
req.environ, method='GET',
headers={'x-auth-token': req.headers.get('x-auth-token')},
agent=('%(orig)s ' + 'SLO MultipartGET'), swift_source='SLO')
resp_iter = self._app_call(get_req.environ)
# Any Content-Range from a manifest is almost certainly wrong for the
# full large object.
resp_headers = [(h, v) for h, v in self._response_headers
if not h.lower() == 'content-range']
response = self.get_or_head_response(
req, resp_headers, resp_iter)
return response(req.environ, start_response)
def handle_slo_get_or_head(self, req, start_response):
"""
Takes a request and a start_response callable and does the normal WSGI
thing with them. Returns an iterator suitable for sending up the WSGI
chain.
:param req: swob.Request object; is a GET or HEAD request aimed at
what may be a static large object manifest (or may not).
:param start_response: WSGI start_response callable
"""
resp_iter = self._app_call(req.environ)
# make sure this response is for a static large object manifest
for header, value in self._response_headers:
if (header.lower() == 'x-static-large-object'
and config_true_value(value)):
break
else:
# Not a static large object manifest. Just pass it through.
start_response(self._response_status, self._response_headers,
self._response_exc_info)
return resp_iter
# Handle pass-through request for the manifest itself
if req.params.get('multipart-manifest') == 'get':
new_headers = []
for header, value in self._response_headers:
if header.lower() == 'content-type':
new_headers.append(
('Content-Type', 'application/json; charset=utf-8'))
else:
new_headers.append((header, value))
self._response_headers = new_headers
start_response(self._response_status, self._response_headers,
self._response_exc_info)
return resp_iter
if self._need_to_refetch_manifest(req):
req.environ['swift.non_client_disconnect'] = True
close_if_possible(resp_iter)
del req.environ['swift.non_client_disconnect']
get_req = make_request(
req.environ,
method='GET',
headers={'x-auth-token': req.headers.get('x-auth-token')},
agent=('%(orig)s ' + 'SLO MultipartGET'),
swift_source='SLO')
resp_iter = self._app_call(get_req.environ)
# Any Content-Range from a manifest is almost certainly wrong for the
# full large object.
resp_headers = [(h, v) for h, v in self._response_headers
if not h.lower() == 'content-range']
response = self.get_or_head_response(req, resp_headers, resp_iter)
return response(req.environ, start_response)
def _get_container_listing(self, req, version, account, container,
prefix, marker=''):
con_req = make_request(
req.environ, path='/'.join(['', version, account, container]),
method='GET',
headers={'x-auth-token': req.headers.get('x-auth-token')},
agent=('%(orig)s ' + 'DLO MultipartGET'), swift_source='DLO')
con_req.query_string = 'format=json&prefix=%s' % quote(prefix)
if marker:
con_req.query_string += '&marker=%s' % quote(marker)
con_resp = con_req.get_response(self.dlo.app)
if not is_success(con_resp.status_int):
return con_resp, None
return None, json.loads(''.join(con_resp.app_iter))
def _get_container_listing(self, req, version, account, container, prefix, marker=""):
con_req = make_request(
req.environ,
path="/".join(["", version, account, container]),
method="GET",
headers={"x-auth-token": req.headers.get("x-auth-token")},
agent=("%(orig)s " + "DLO MultipartGET"),
swift_source="DLO",
)
con_req.query_string = "format=json&prefix=%s" % quote(prefix)
if marker:
con_req.query_string += "&marker=%s" % quote(marker)
con_resp = con_req.get_response(self.dlo.app)
if not is_success(con_resp.status_int):
return con_resp, None
return None, json.loads("".join(con_resp.app_iter))
def __iter__(self):
start_time = time.time()
have_yielded_data = False
if self.response and self.response.content_length:
bytes_left = int(self.response.content_length)
else:
bytes_left = None
try:
for seg_path, seg_etag, seg_size, first_byte, last_byte in self.listing_iter:
if time.time() - start_time > self.max_get_time:
raise SegmentError(
"ERROR: While processing manifest %s, "
"max LO GET time of %ds exceeded" % (self.name, self.max_get_time)
)
seg_req = make_request(
self.req.environ,
path=seg_path,
method="GET",
headers={"x-auth-token": self.req.headers.get("x-auth-token")},
agent=("%(orig)s " + self.ua_suffix),
swift_source=self.swift_source,
)
if first_byte is not None or last_byte is not None:
seg_req.headers["Range"] = "bytes=%s-%s" % (
# The 0 is to avoid having a range like "bytes=-10",
# which actually means the *last* 10 bytes.
"0" if first_byte is None else first_byte,
"" if last_byte is None else last_byte,
)
seg_resp = seg_req.get_response(self.app)
if not is_success(seg_resp.status_int):
close_if_possible(seg_resp.app_iter)
raise SegmentError(
"ERROR: While processing manifest %s, "
"got %d while retrieving %s" % (self.name, seg_resp.status_int, seg_path)
)
elif (seg_etag and (seg_resp.etag != seg_etag)) or (
seg_size and (seg_resp.content_length != seg_size) and not seg_req.range
):
# The content-length check is for security reasons. Seems
# possible that an attacker could upload a >1mb object and
# then replace it with a much smaller object with same
# etag. Then create a big nested SLO that calls that
# object many times which would hammer our obj servers. If
# this is a range request, don't check content-length
# because it won't match.
close_if_possible(seg_resp.app_iter)
raise SegmentError(
"Object segment no longer valid: "
"%(path)s etag: %(r_etag)s != %(s_etag)s or "
"%(r_size)s != %(s_size)s."
% {
"path": seg_req.path,
"r_etag": seg_resp.etag,
"r_size": seg_resp.content_length,
"s_etag": seg_etag,
"s_size": seg_size,
}
)
for chunk in seg_resp.app_iter:
have_yielded_data = True
if bytes_left is None:
yield chunk
elif bytes_left >= len(chunk):
yield chunk
bytes_left -= len(chunk)
else:
yield chunk[:bytes_left]
bytes_left -= len(chunk)
close_if_possible(seg_resp.app_iter)
raise SegmentError(
"Too many bytes for %(name)s; truncating in "
"%(seg)s with %(left)d bytes left"
% {"name": self.name, "seg": seg_req.path, "left": bytes_left}
)
close_if_possible(seg_resp.app_iter)
if bytes_left:
raise SegmentError("Not enough bytes for %s; closing connection" % self.name)
except ListingIterError as err:
# I have to save this error because yielding the ' ' below clears
# the exception from the current stack frame.
excinfo = sys.exc_info()
self.logger.exception("ERROR: While processing manifest %s, %s", self.name, err)
# Normally, exceptions before any data has been yielded will
# cause Eventlet to send a 5xx response. In this particular
# case of ListingIterError we don't want that and we'd rather
# just send the normal 2xx response and then hang up early
# since 5xx codes are often used to judge Service Level
# Agreements and this ListingIterError indicates the user has
# created an invalid condition.
if not have_yielded_data:
yield " "
raise excinfo
except SegmentError as err:
self.logger.exception(err)
# This doesn't actually change the response status (we're too
# late for that), but this does make it to the logs.
if self.response:
self.response.status = HTTP_SERVICE_UNAVAILABLE
raise
def __iter__(self):
start_time = time.time()
have_yielded_data = False
if self.response and self.response.content_length:
bytes_left = int(self.response.content_length)
else:
bytes_left = None
try:
for seg_path, seg_etag, seg_size, first_byte, last_byte \
in self.listing_iter:
if time.time() - start_time > self.max_get_time:
raise SegmentError('ERROR: While processing manifest %s, '
'max LO GET time of %ds exceeded' %
(self.name, self.max_get_time))
seg_req = make_request(self.req.environ,
path=seg_path,
method='GET',
headers={
'x-auth-token':
self.req.headers.get('x-auth-token')
},
agent=('%(orig)s ' + self.ua_suffix),
swift_source=self.swift_source)
if first_byte is not None or last_byte is not None:
seg_req.headers['Range'] = "bytes=%s-%s" % (
# The 0 is to avoid having a range like "bytes=-10",
# which actually means the *last* 10 bytes.
'0' if first_byte is None else first_byte,
'' if last_byte is None else last_byte)
seg_resp = seg_req.get_response(self.app)
if not is_success(seg_resp.status_int):
close_if_possible(seg_resp.app_iter)
raise SegmentError(
'ERROR: While processing manifest %s, '
'got %d while retrieving %s' %
(self.name, seg_resp.status_int, seg_path))
elif ((seg_etag and (seg_resp.etag != seg_etag))
or (seg_size and (seg_resp.content_length != seg_size)
and not seg_req.range)):
# The content-length check is for security reasons. Seems
# possible that an attacker could upload a >1mb object and
# then replace it with a much smaller object with same
# etag. Then create a big nested SLO that calls that
# object many times which would hammer our obj servers. If
# this is a range request, don't check content-length
# because it won't match.
close_if_possible(seg_resp.app_iter)
raise SegmentError(
'Object segment no longer valid: '
'%(path)s etag: %(r_etag)s != %(s_etag)s or '
'%(r_size)s != %(s_size)s.' % {
'path': seg_req.path,
'r_etag': seg_resp.etag,
'r_size': seg_resp.content_length,
's_etag': seg_etag,
's_size': seg_size
})
for chunk in seg_resp.app_iter:
have_yielded_data = True
if bytes_left is None:
yield chunk
elif bytes_left >= len(chunk):
yield chunk
bytes_left -= len(chunk)
else:
yield chunk[:bytes_left]
bytes_left -= len(chunk)
close_if_possible(seg_resp.app_iter)
raise SegmentError(
'Too many bytes for %(name)s; truncating in '
'%(seg)s with %(left)d bytes left' % {
'name': self.name,
'seg': seg_req.path,
'left': bytes_left
})
close_if_possible(seg_resp.app_iter)
if bytes_left:
raise SegmentError(
'Not enough bytes for %s; closing connection' % self.name)
except ListingIterError as err:
# I have to save this error because yielding the ' ' below clears
# the exception from the current stack frame.
excinfo = sys.exc_info()
self.logger.exception('ERROR: While processing manifest %s, %s',
self.name, err)
# Normally, exceptions before any data has been yielded will
# cause Eventlet to send a 5xx response. In this particular
# case of ListingIterError we don't want that and we'd rather
# just send the normal 2xx response and then hang up early
# since 5xx codes are often used to judge Service Level
# Agreements and this ListingIterError indicates the user has
# created an invalid condition.
if not have_yielded_data:
yield ' '
raise excinfo
except SegmentError as err:
self.logger.exception(err)
# This doesn't actually change the response status (we're too
# late for that), but this does make it to the logs.
if self.response:
self.response.status = HTTP_SERVICE_UNAVAILABLE
raise