quay/endpoints/v2/blob.py

import logging
import re

from flask import make_response, url_for, request, redirect, Response, abort as flask_abort

from app import storage, app
from data import model, database
from digest import digest_tools
from endpoints.v2 import v2_bp, require_repo_read, require_repo_write, get_input_stream
from endpoints.v2.errors import BlobUnknown, BlobUploadInvalid, BlobUploadUnknown, Unsupported
from auth.jwt_auth import process_jwt_auth
from endpoints.decorators import anon_protect
from util.cache import cache_control
from util.registry.filelike import wrap_with_handler, StreamSlice
from storage.basestorage import InvalidChunkException


logger = logging.getLogger(__name__)


BASE_BLOB_ROUTE = '/<namespace>/<repo_name>/blobs/<regex("{0}"):digest>'
BLOB_DIGEST_ROUTE = BASE_BLOB_ROUTE.format(digest_tools.DIGEST_PATTERN)
RANGE_HEADER_REGEX = re.compile(r'^bytes=([0-9]+)-([0-9]+)$')


class _InvalidRangeHeader(Exception):
  pass


def _base_blob_fetch(namespace, repo_name, digest):
  """ Some work that is common to both GET and HEAD requests. Callers MUST check for proper
      authorization before calling this method.
  """
  try:
    found = model.blob.get_repo_blob_by_digest(namespace, repo_name, digest)
  except model.BlobDoesNotExist:
    raise BlobUnknown()

  headers = {
    'Docker-Content-Digest': digest,
  }

  # Add the Accept-Ranges header if the storage engine supports resumable
  # downloads.
  if storage.get_supports_resumable_downloads(found.locations):
    logger.debug('Storage supports resumable downloads')
    headers['Accept-Ranges'] = 'bytes'

  return found, headers


@v2_bp.route(BLOB_DIGEST_ROUTE, methods=['HEAD'])
@process_jwt_auth
@require_repo_read
@anon_protect
@cache_control(max_age=31436000)
def check_blob_exists(namespace, repo_name, digest):
  found, headers = _base_blob_fetch(namespace, repo_name, digest)

  response = make_response('')
  response.headers.extend(headers)
  response.headers['Content-Length'] = found.image_size
  return response


@v2_bp.route(BLOB_DIGEST_ROUTE, methods=['GET'])
@process_jwt_auth
@require_repo_read
@anon_protect
@cache_control(max_age=31536000)
def download_blob(namespace, repo_name, digest):
  found, headers = _base_blob_fetch(namespace, repo_name, digest)

  path = model.storage.get_layer_path(found)
  logger.debug('Looking up the direct download URL for path: %s', path)
  direct_download_url = storage.get_direct_download_url(found.locations, path)

  if direct_download_url:
    logger.debug('Returning direct download URL')
    resp = redirect(direct_download_url)
    resp.headers.extend(headers)
    return resp

  logger.debug('Streaming layer data')

  # Close the database handle here for this process before we send the long download.
  database.close_db_filter(None)

  headers['Content-Length'] = found.image_size

  return Response(storage.stream_read(found.locations, path), headers=headers)


def _render_range(num_uploaded_bytes, with_bytes_prefix=True):
  return '{0}0-{1}'.format('bytes=' if with_bytes_prefix else '', num_uploaded_bytes - 1)


@v2_bp.route('/<namespace>/<repo_name>/blobs/uploads/', methods=['POST'])
@process_jwt_auth
@require_repo_write
@anon_protect
def start_blob_upload(namespace, repo_name):
  location_name = storage.preferred_locations[0]
  new_upload_uuid, upload_metadata = storage.initiate_chunked_upload(location_name)
  model.blob.initiate_upload(namespace, repo_name, new_upload_uuid, location_name, upload_metadata)

  digest = request.args.get('digest', None)
  if digest is None:
    # The user will send the blob data in another request
    accepted = make_response('', 202)
    accepted.headers['Location'] = url_for('v2.upload_chunk', namespace=namespace,
                                           repo_name=repo_name, upload_uuid=new_upload_uuid)
    accepted.headers['Range'] = _render_range(0)
    accepted.headers['Docker-Upload-UUID'] = new_upload_uuid
    return accepted
  else:
    # The user plans to send us the entire body right now
    uploaded = _upload_chunk(namespace, repo_name, new_upload_uuid)
    uploaded.save()

    return _finish_upload(namespace, repo_name, uploaded, digest)


@v2_bp.route('/<namespace>/<repo_name>/blobs/uploads/<upload_uuid>', methods=['GET'])
@process_jwt_auth
@require_repo_write
@anon_protect
def fetch_existing_upload(namespace, repo_name, upload_uuid):
  try:
    found = model.blob.get_blob_upload(namespace, repo_name, upload_uuid)
  except model.InvalidBlobUpload:
    raise BlobUploadUnknown()

  # Note: Docker byte ranges are exclusive so we have to add one to the byte count.
  accepted = make_response('', 204)
  accepted.headers['Range'] = _render_range(found.byte_count + 1)
  accepted.headers['Docker-Upload-UUID'] = upload_uuid
  return accepted


def _current_request_path():
  return '{0}{1}'.format(request.script_root, request.path)


def _range_not_satisfiable(valid_end):
  invalid_range = make_response('', 416)
  invalid_range.headers['Location'] = _current_request_path()
  invalid_range.headers['Range'] = '0-{0}'.format(valid_end)
  invalid_range.headers['Docker-Upload-UUID'] = request.view_args['upload_uuid']
  flask_abort(invalid_range)


def _parse_range_header(range_header_text):
  """ Parses the range header, and returns a tuple of the start offset and the length,
      or raises an _InvalidRangeHeader exception.
  """
  found = RANGE_HEADER_REGEX.match(range_header_text)
  if found is None:
    raise _InvalidRangeHeader()

  start = int(found.group(1))
  length = int(found.group(2)) - start

  if length <= 0:
    raise _InvalidRangeHeader()

  return (start, length)


def _upload_chunk(namespace, repo_name, upload_uuid):
  """ Common code among the various uploading paths for appending data to blobs.
      Callers MUST call .save() or .delete_instance() on the returned database object.
  """
  try:
    found = model.blob.get_blob_upload(namespace, repo_name, upload_uuid)
  except model.InvalidBlobUpload:
    raise BlobUploadUnknown()

  start_offset, length = 0, -1
  range_header = request.headers.get('range', None)
  if range_header is not None:
    try:
      start_offset, length = _parse_range_header(range_header)
    except _InvalidRangeHeader:
      _range_not_satisfiable(found.byte_count)

  if start_offset > 0 and start_offset > found.byte_count:
    _range_not_satisfiable(found.byte_count)

  input_fp = get_input_stream(request)

  if start_offset > 0 and start_offset < found.byte_count:
    # Skip the bytes which were received on a previous push, which are already stored and
    # included in the sha calculation
    input_fp = StreamSlice(input_fp, found.byte_count - start_offset)
    start_offset = found.byte_count

  input_fp = wrap_with_handler(input_fp, found.sha_state.update)

  try:
    length_written, new_metadata = storage.stream_upload_chunk({found.location.name}, upload_uuid,
                                                               start_offset, length, input_fp,
                                                               found.storage_metadata)
  except InvalidChunkException:
    _range_not_satisfiable(found.byte_count)

  found.storage_metadata = new_metadata
  found.byte_count += length_written
  return found


def _finish_upload(namespace, repo_name, upload_obj, expected_digest):
  # Verify that the digest's SHA matches that of the uploaded data.
  computed_digest = digest_tools.sha256_digest_from_hashlib(upload_obj.sha_state)
  if not digest_tools.digests_equal(computed_digest, expected_digest):
    raise BlobUploadInvalid()

  final_blob_location = digest_tools.content_path(expected_digest)

  # Move the storage into place, or if this was a re-upload, cancel it
  with database.CloseForLongOperation(app.config):
    if storage.exists({upload_obj.location.name}, final_blob_location):
      # It already existed, clean up our upload which served as proof that we had the file
      storage.cancel_chunked_upload({upload_obj.location.name}, upload_obj.uuid,
                                    upload_obj.storage_metadata)

    else:
      # We were the first ones to upload this image (at least to this location)
      # Let's copy it into place
      storage.complete_chunked_upload({upload_obj.location.name}, upload_obj.uuid,
                                      final_blob_location, upload_obj.storage_metadata)

  # Mark the blob as uploaded.
  model.blob.store_blob_record_and_temp_link(namespace, repo_name, expected_digest,
                                             upload_obj.location, upload_obj.byte_count,
                                             app.config['PUSH_TEMP_TAG_EXPIRATION_SEC'])

  # Delete the upload tracking row.
  upload_obj.delete_instance()

  response = make_response('', 201)
  response.headers['Docker-Content-Digest'] = expected_digest
  response.headers['Location'] = url_for('v2.download_blob', namespace=namespace,
                                         repo_name=repo_name, digest=expected_digest)
  return response


@v2_bp.route('/<namespace>/<repo_name>/blobs/uploads/<upload_uuid>', methods=['PATCH'])
@process_jwt_auth
@require_repo_write
@anon_protect
def upload_chunk(namespace, repo_name, upload_uuid):
  upload = _upload_chunk(namespace, repo_name, upload_uuid)
  upload.save()

  accepted = make_response('', 204)
  accepted.headers['Location'] = _current_request_path()
  accepted.headers['Range'] = _render_range(upload.byte_count, with_bytes_prefix=False)
  accepted.headers['Docker-Upload-UUID'] = upload_uuid
  return accepted


@v2_bp.route('/<namespace>/<repo_name>/blobs/uploads/<upload_uuid>', methods=['PUT'])
@process_jwt_auth
@require_repo_write
@anon_protect
def monolithic_upload_or_last_chunk(namespace, repo_name, upload_uuid):
  digest = request.args.get('digest', None)
  if digest is None:
    raise BlobUploadInvalid()

  found = _upload_chunk(namespace, repo_name, upload_uuid)
  return _finish_upload(namespace, repo_name, found, digest)


@v2_bp.route('/<namespace>/<repo_name>/blobs/uploads/<upload_uuid>', methods=['DELETE'])
@process_jwt_auth
@require_repo_write
@anon_protect
def cancel_upload(namespace, repo_name, upload_uuid):
  try:
    found = model.blob.get_blob_upload(namespace, repo_name, upload_uuid)
  except model.InvalidBlobUpload:
    raise BlobUploadUnknown()

  # We delete the record for the upload first, since if the partial upload in
  # storage fails to delete, it doesn't break anything
  found.delete_instance()
  storage.cancel_chunked_upload({found.location.name}, found.uuid, found.storage_metadata)

  return make_response('', 204)



@v2_bp.route('/<namespace>/<repo_name>/blobs/<digest>', methods=['DELETE'])
@process_jwt_auth
@require_repo_write
@anon_protect
def delete_digest(namespace, repo_name, upload_uuid):
  # We do not support deleting arbitrary digests, as they break repo images.
  raise Unsupported()