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_and_clear_replication_confirmed
14 before_validation :set_portable_data_hash
15 validate :ensure_hash_matches_manifest_text
16 before_save :set_file_names
18 # Query only undeleted collections by default.
19 default_scope where("expires_at IS NULL or expires_at > CURRENT_TIMESTAMP")
21 api_accessible :user, extend: :common do |t|
25 t.add :portable_data_hash
26 t.add :signed_manifest_text, as: :manifest_text
27 t.add :replication_desired
28 t.add :replication_confirmed
29 t.add :replication_confirmed_at
32 LOCATOR_REGEXP = /^([[:xdigit:]]{32})(\+([[:digit:]]+))?(\+([[:upper:]][[:alnum:]+@_-]*))?$/
34 def self.attributes_required_columns
36 # If we don't list manifest_text explicitly, the
37 # params[:select] code gets confused by the way we
38 # expose signed_manifest_text as manifest_text in the
39 # API response, and never let clients select the
40 # manifest_text column.
41 'manifest_text' => ['manifest_text'],
46 return false if self.manifest_text.nil?
48 return true if current_user.andand.is_admin
50 # Provided the manifest_text hasn't changed materially since an
51 # earlier validation, it's safe to pass this validation on
52 # subsequent passes without checking any signatures. This is
53 # important because the signatures have probably been stripped off
54 # by the time we get to a second validation pass!
55 computed_pdh = compute_pdh
56 return true if @signatures_checked and @signatures_checked == computed_pdh
58 if self.manifest_text_changed?
59 # Check permissions on the collection manifest.
60 # If any signature cannot be verified, raise PermissionDeniedError
61 # which will return 403 Permission denied to the client.
62 api_token = current_api_client_authorization.andand.api_token
65 now: db_current_time.to_i,
67 self.manifest_text.lines.each do |entry|
68 entry.split[1..-1].each do |tok|
69 if /^[[:digit:]]+:[[:digit:]]+:/.match tok
70 # This is a filename token, not a blob locator. Note that we
71 # keep checking tokens after this, even though manifest
72 # format dictates that all subsequent tokens will also be
73 # filenames. Safety first!
74 elsif Blob.verify_signature tok, signing_opts
76 elsif Keep::Locator.parse(tok).andand.signature
77 # Signature provided, but verify_signature did not like it.
78 logger.warn "Invalid signature on locator #{tok}"
79 raise ArvadosModel::PermissionDeniedError
80 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
81 # No signature provided, but we are running in insecure mode.
82 logger.debug "Missing signature on locator #{tok} ignored"
83 elsif Blob.new(tok).empty?
84 # No signature provided -- but no data to protect, either.
86 logger.warn "Missing signature on locator #{tok}"
87 raise ArvadosModel::PermissionDeniedError
92 @signatures_checked = computed_pdh
95 def strip_manifest_text_and_clear_replication_confirmed
96 if self.manifest_text_changed?
98 self.class.munge_manifest_locators!(manifest_text_was) do |match|
99 in_old_manifest[match[1]] = true
102 cleared_replication_confirmed = false
104 # Remove any permission signatures from the manifest.
105 self[:manifest_text] = self.class.munge_manifest_locators!(self[:manifest_text]) do |match|
106 if not cleared_replication_confirmed and not in_old_manifest[match[1]]
107 self.replication_confirmed_at = nil
108 self.replication_confirmed = nil
109 cleared_replication_confirmed = true
111 self.class.locator_without_signature(match)
117 def self.locator_without_signature match
118 without_signature = match[1]
119 without_signature += match[2] if match[2]
121 hints = match[4].split('+').reject { |hint| hint.start_with?("A") }
122 without_signature += hints.join('+')
127 def set_portable_data_hash
128 if (portable_data_hash.nil? or
129 portable_data_hash == "" or
130 (manifest_text_changed? and !portable_data_hash_changed?))
131 @need_pdh_validation = false
132 self.portable_data_hash = compute_pdh
133 elsif portable_data_hash_changed?
134 @need_pdh_validation = true
136 loc = Keep::Locator.parse!(self.portable_data_hash)
139 self.portable_data_hash = loc.to_s
141 self.portable_data_hash = "#{loc.hash}+#{portable_manifest_text.bytesize}"
143 rescue ArgumentError => e
144 errors.add(:portable_data_hash, "#{e}")
151 def ensure_hash_matches_manifest_text
152 return true unless manifest_text_changed? or portable_data_hash_changed?
153 # No need verify it if :set_portable_data_hash just computed it!
154 return true if not @need_pdh_validation
155 expect_pdh = compute_pdh
156 if expect_pdh != portable_data_hash
157 errors.add(:portable_data_hash,
158 "does not match computed hash #{expect_pdh}")
164 if self.manifest_text_changed?
165 self.file_names = manifest_files
172 if self.manifest_text
173 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
174 names << name.first.gsub('\040',' ') + "\n"
175 break if names.length > 2**12
179 if self.manifest_text and names.length < 2**12
180 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
181 names << stream_name.first.gsub('\040',' ') + "\n"
182 break if names.length > 2**12
190 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
194 # If Ruby thinks the encoding is something else, like 7-bit
195 # ASCII, but its stored bytes are equal to the (valid) UTF-8
196 # encoding of the same string, we declare it to be a UTF-8
199 utf8.force_encoding Encoding::UTF_8
200 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
206 errors.add :manifest_text, "must use UTF-8 encoding"
211 def signed_manifest_text
212 if has_attribute? :manifest_text
213 token = current_api_client_authorization.andand.api_token
214 @signed_manifest_text = self.class.sign_manifest manifest_text, token
218 def self.sign_manifest manifest, token
221 expire: db_current_time.to_i + Rails.configuration.blob_signature_ttl,
224 m = munge_manifest_locators!(m) do |match|
225 Blob.sign_locator(locator_without_signature(match), signing_opts)
230 def self.munge_manifest_locators! manifest
231 # Given a manifest text and a block, yield each locator,
232 # and replace it with whatever the block returns.
234 lines = manifest.andand.split("\n")
235 lines.andand.each do |line|
236 words = line.split(' ')
239 if match = LOCATOR_REGEXP.match(word.strip)
240 new_words << yield(match)
242 new_words << word.strip
245 new_lines << new_words.join(' ')
249 ends_with_newline = manifest.end_with?("\n")
250 manifest = new_lines.join("\n")
251 manifest += "\n" if ends_with_newline
257 def self.normalize_uuid uuid
260 uuid.split('+').each do |token|
261 if token.match /^[0-9a-f]{32,}$/
262 raise "uuid #{uuid} has multiple hash parts" if hash_part
264 elsif token.match /^\d+$/
265 raise "uuid #{uuid} has multiple size parts" if size_part
269 raise "uuid #{uuid} has no hash part" if !hash_part
270 [hash_part, size_part].compact.join '+'
273 # Return array of Collection objects
274 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
275 readers ||= [Thread.current[:user]]
277 readable_by(*readers).
278 readable_by(*readers, table_name: "collections").
279 joins("JOIN collections ON links.head_uuid = collections.uuid").
280 order("links.created_at DESC")
282 # If the search term is a Collection locator that contains one file
283 # that looks like a Docker image, return it.
284 if loc = Keep::Locator.parse(search_term)
286 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
288 # Check if the Collection contains exactly one file whose name
289 # looks like a saved Docker image.
290 manifest = Keep::Manifest.new(coll_match.manifest_text)
291 if manifest.exact_file_count?(1) and
292 (manifest.files[0][1] =~ /^[0-9A-Fa-f]{64}\.tar$/)
298 if search_tag.nil? and (n = search_term.index(":"))
299 search_tag = search_term[n+1..-1]
300 search_term = search_term[0..n-1]
303 # Find Collections with matching Docker image repository+tag pairs.
304 matches = base_search.
305 where(link_class: "docker_image_repo+tag",
306 name: "#{search_term}:#{search_tag || 'latest'}")
308 # If that didn't work, find Collections with matching Docker image hashes.
310 matches = base_search.
311 where("link_class = ? and links.name LIKE ?",
312 "docker_image_hash", "#{search_term}%")
315 # Generate an order key for each result. We want to order the results
316 # so that anything with an image timestamp is considered more recent than
317 # anything without; then we use the link's created_at as a tiebreaker.
319 matches.all.map do |link|
320 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
321 -link.created_at.to_i]
323 Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
326 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
327 find_all_for_docker_image(search_term, search_tag, readers).first
330 def self.searchable_columns operator
331 super - ["manifest_text"]
334 def self.full_text_searchable_columns
335 super - ["manifest_text"]
339 def portable_manifest_text
340 portable_manifest = self[:manifest_text].dup
341 portable_manifest = self.class.munge_manifest_locators!(portable_manifest) do |match|
352 return @computed_pdh if @computed_pdh
353 portable_manifest = portable_manifest_text
354 @computed_pdh = (Digest::MD5.hexdigest(portable_manifest) +
356 portable_manifest.bytesize.to_s)
360 def ensure_permission_to_save
361 if (not current_user.andand.is_admin and
362 (replication_confirmed_at_changed? or replication_confirmed_changed?) and
363 not (replication_confirmed_at.nil? and replication_confirmed.nil?))
364 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")