3 class Collection < ArvadosModel
7 include CommonApiTemplate
9 serialize :properties, Hash
11 before_validation :check_encoding
12 before_validation :check_signatures
13 before_validation :strip_manifest_text
14 before_validation :set_portable_data_hash
15 before_validation :maybe_clear_replication_confirmed
16 validate :ensure_hash_matches_manifest_text
17 before_save :set_file_names
19 # Query only undeleted collections by default.
20 default_scope where("expires_at IS NULL or expires_at > CURRENT_TIMESTAMP")
22 api_accessible :user, extend: :common do |t|
26 t.add :portable_data_hash
27 t.add :signed_manifest_text, as: :manifest_text
28 t.add :replication_desired
29 t.add :replication_confirmed
30 t.add :replication_confirmed_at
33 def self.attributes_required_columns
35 # If we don't list manifest_text explicitly, the
36 # params[:select] code gets confused by the way we
37 # expose signed_manifest_text as manifest_text in the
38 # API response, and never let clients select the
39 # manifest_text column.
40 'manifest_text' => ['manifest_text'],
45 return false if self.manifest_text.nil?
47 return true if current_user.andand.is_admin
49 # Provided the manifest_text hasn't changed materially since an
50 # earlier validation, it's safe to pass this validation on
51 # subsequent passes without checking any signatures. This is
52 # important because the signatures have probably been stripped off
53 # by the time we get to a second validation pass!
54 return true if @signatures_checked and @signatures_checked == compute_pdh
56 if self.manifest_text_changed?
57 # Check permissions on the collection manifest.
58 # If any signature cannot be verified, raise PermissionDeniedError
59 # which will return 403 Permission denied to the client.
60 api_token = current_api_client_authorization.andand.api_token
63 now: db_current_time.to_i,
65 self.manifest_text.lines.each do |entry|
66 entry.split[1..-1].each do |tok|
67 if /^[[:digit:]]+:[[:digit:]]+:/.match tok
68 # This is a filename token, not a blob locator. Note that we
69 # keep checking tokens after this, even though manifest
70 # format dictates that all subsequent tokens will also be
71 # filenames. Safety first!
72 elsif Blob.verify_signature tok, signing_opts
74 elsif Keep::Locator.parse(tok).andand.signature
75 # Signature provided, but verify_signature did not like it.
76 logger.warn "Invalid signature on locator #{tok}"
77 raise ArvadosModel::PermissionDeniedError
78 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
79 # No signature provided, but we are running in insecure mode.
80 logger.debug "Missing signature on locator #{tok} ignored"
81 elsif Blob.new(tok).empty?
82 # No signature provided -- but no data to protect, either.
84 logger.warn "Missing signature on locator #{tok}"
85 raise ArvadosModel::PermissionDeniedError
90 @signatures_checked = compute_pdh
93 def strip_manifest_text
94 if self.manifest_text_changed?
95 # Remove any permission signatures from the manifest.
96 self.class.munge_manifest_locators!(self[:manifest_text]) do |loc|
97 loc.without_signature.to_s
103 def set_portable_data_hash
104 if (portable_data_hash.nil? or
105 portable_data_hash == "" or
106 (manifest_text_changed? and !portable_data_hash_changed?))
107 @need_pdh_validation = false
108 self.portable_data_hash = compute_pdh
109 elsif portable_data_hash_changed?
110 @need_pdh_validation = true
112 loc = Keep::Locator.parse!(self.portable_data_hash)
115 self.portable_data_hash = loc.to_s
117 self.portable_data_hash = "#{loc.hash}+#{portable_manifest_text.bytesize}"
119 rescue ArgumentError => e
120 errors.add(:portable_data_hash, "#{e}")
127 def ensure_hash_matches_manifest_text
128 return true unless manifest_text_changed? or portable_data_hash_changed?
129 # No need verify it if :set_portable_data_hash just computed it!
130 return true if not @need_pdh_validation
131 expect_pdh = compute_pdh
132 if expect_pdh != portable_data_hash
133 errors.add(:portable_data_hash,
134 "does not match computed hash #{expect_pdh}")
140 if self.manifest_text_changed?
141 self.file_names = manifest_files
148 if self.manifest_text
149 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
150 names << name.first.gsub('\040',' ') + "\n"
151 break if names.length > 2**12
155 if self.manifest_text and names.length < 2**12
156 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
157 names << stream_name.first.gsub('\040',' ') + "\n"
158 break if names.length > 2**12
166 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
170 # If Ruby thinks the encoding is something else, like 7-bit
171 # ASCII, but its stored bytes are equal to the (valid) UTF-8
172 # encoding of the same string, we declare it to be a UTF-8
175 utf8.force_encoding Encoding::UTF_8
176 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
182 errors.add :manifest_text, "must use UTF-8 encoding"
187 def signed_manifest_text
188 if has_attribute? :manifest_text
189 token = current_api_client_authorization.andand.api_token
190 @signed_manifest_text = self.class.sign_manifest manifest_text, token
194 def self.sign_manifest manifest, token
197 expire: db_current_time.to_i + Rails.configuration.blob_signature_ttl,
200 munge_manifest_locators!(m) do |loc|
201 Blob.sign_locator(loc.to_s, signing_opts)
206 def self.munge_manifest_locators! manifest
207 # Given a manifest text and a block, yield each locator,
208 # and replace it with whatever the block returns.
209 manifest.andand.gsub!(/ [[:xdigit:]]{32}(\+\S+)?/) do |word|
210 if loc = Keep::Locator.parse(word.strip)
218 def self.each_manifest_locator manifest
219 # Given a manifest text and a block, yield each locator.
220 manifest.andand.scan(/ ([[:xdigit:]]{32}(\+\S+)?)/) do |word, _|
221 if loc = Keep::Locator.parse(word)
227 def self.normalize_uuid uuid
230 uuid.split('+').each do |token|
231 if token.match /^[0-9a-f]{32,}$/
232 raise "uuid #{uuid} has multiple hash parts" if hash_part
234 elsif token.match /^\d+$/
235 raise "uuid #{uuid} has multiple size parts" if size_part
239 raise "uuid #{uuid} has no hash part" if !hash_part
240 [hash_part, size_part].compact.join '+'
243 # Return array of Collection objects
244 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
245 readers ||= [Thread.current[:user]]
247 readable_by(*readers).
248 readable_by(*readers, table_name: "collections").
249 joins("JOIN collections ON links.head_uuid = collections.uuid").
250 order("links.created_at DESC")
252 # If the search term is a Collection locator that contains one file
253 # that looks like a Docker image, return it.
254 if loc = Keep::Locator.parse(search_term)
256 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
258 # Check if the Collection contains exactly one file whose name
259 # looks like a saved Docker image.
260 manifest = Keep::Manifest.new(coll_match.manifest_text)
261 if manifest.exact_file_count?(1) and
262 (manifest.files[0][1] =~ /^[0-9A-Fa-f]{64}\.tar$/)
268 if search_tag.nil? and (n = search_term.index(":"))
269 search_tag = search_term[n+1..-1]
270 search_term = search_term[0..n-1]
273 # Find Collections with matching Docker image repository+tag pairs.
274 matches = base_search.
275 where(link_class: "docker_image_repo+tag",
276 name: "#{search_term}:#{search_tag || 'latest'}")
278 # If that didn't work, find Collections with matching Docker image hashes.
280 matches = base_search.
281 where("link_class = ? and links.name LIKE ?",
282 "docker_image_hash", "#{search_term}%")
285 # Generate an order key for each result. We want to order the results
286 # so that anything with an image timestamp is considered more recent than
287 # anything without; then we use the link's created_at as a tiebreaker.
289 matches.all.map do |link|
290 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
291 -link.created_at.to_i]
293 Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
296 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
297 find_all_for_docker_image(search_term, search_tag, readers).first
300 def self.searchable_columns operator
301 super - ["manifest_text"]
304 def self.full_text_searchable_columns
305 super - ["manifest_text"]
309 def portable_manifest_text
310 portable_manifest = self[:manifest_text].dup
311 self.class.munge_manifest_locators!(portable_manifest) do |loc|
313 loc.hash + '+' + loc.size.to_s
322 portable_manifest = portable_manifest_text
323 (Digest::MD5.hexdigest(portable_manifest) +
325 portable_manifest.bytesize.to_s)
328 def maybe_clear_replication_confirmed
329 if manifest_text_changed?
330 # If the new manifest_text contains locators whose hashes
331 # weren't in the old manifest_text, storage replication is no
334 self.class.each_manifest_locator(manifest_text_was) do |loc|
335 in_old_manifest[loc.hash] = true
337 self.class.each_manifest_locator(manifest_text) do |loc|
338 if not in_old_manifest[loc.hash]
339 self.replication_confirmed_at = nil
340 self.replication_confirmed = nil
347 def ensure_permission_to_save
348 if (not current_user.andand.is_admin and
349 (replication_confirmed_at_changed? or replication_confirmed_changed?) and
350 not (replication_confirmed_at.nil? and replication_confirmed.nil?))
351 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")