1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: AGPL-3.0
6 require 'sweep_trashed_collections'
9 class Collection < ArvadosModel
10 extend CurrentApiClient
14 include CommonApiTemplate
17 serialize :properties, Hash
18 serialize :storage_classes_desired, Array
19 serialize :storage_classes_confirmed, Array
21 before_validation :default_empty_manifest
22 before_validation :check_encoding
23 before_validation :check_manifest_validity
24 before_validation :check_signatures
25 before_validation :strip_signatures_and_update_replication_confirmed
26 validate :ensure_pdh_matches_manifest_text
27 validate :ensure_storage_classes_desired_is_not_empty
28 validate :ensure_storage_classes_contain_non_empty_strings
29 before_save :set_file_names
31 api_accessible :user, extend: :common do |t|
35 t.add :portable_data_hash
36 t.add :signed_manifest_text, as: :manifest_text
37 t.add :manifest_text, as: :unsigned_manifest_text
38 t.add :replication_desired
39 t.add :replication_confirmed
40 t.add :replication_confirmed_at
41 t.add :storage_classes_desired
42 t.add :storage_classes_confirmed
43 t.add :storage_classes_confirmed_at
50 @signatures_checked = false
51 @computed_pdh_for_manifest_text = false
54 def self.attributes_required_columns
56 # If we don't list manifest_text explicitly, the
57 # params[:select] code gets confused by the way we
58 # expose signed_manifest_text as manifest_text in the
59 # API response, and never let clients select the
60 # manifest_text column.
62 # We need trash_at and is_trashed to determine the
63 # correct timestamp in signed_manifest_text.
64 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
65 'unsigned_manifest_text' => ['manifest_text'],
69 def self.ignored_select_attributes
70 super + ["updated_at", "file_names"]
73 def self.limit_index_columns_read
77 FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
79 return false if self.manifest_text.nil?
81 return true if current_user.andand.is_admin
83 # Provided the manifest_text hasn't changed materially since an
84 # earlier validation, it's safe to pass this validation on
85 # subsequent passes without checking any signatures. This is
86 # important because the signatures have probably been stripped off
87 # by the time we get to a second validation pass!
88 if @signatures_checked && @signatures_checked == computed_pdh
92 if self.manifest_text_changed?
93 # Check permissions on the collection manifest.
94 # If any signature cannot be verified, raise PermissionDeniedError
95 # which will return 403 Permission denied to the client.
96 api_token = current_api_client_authorization.andand.api_token
99 now: @validation_timestamp.to_i,
101 self.manifest_text.each_line do |entry|
102 entry.split.each do |tok|
103 if tok == '.' or tok.starts_with? './'
105 elsif tok =~ FILE_TOKEN
106 # This is a filename token, not a blob locator. Note that we
107 # keep checking tokens after this, even though manifest
108 # format dictates that all subsequent tokens will also be
109 # filenames. Safety first!
110 elsif Blob.verify_signature tok, signing_opts
112 elsif Keep::Locator.parse(tok).andand.signature
113 # Signature provided, but verify_signature did not like it.
114 logger.warn "Invalid signature on locator #{tok}"
115 raise ArvadosModel::PermissionDeniedError
116 elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
117 # No signature provided, but we are running in insecure mode.
118 logger.debug "Missing signature on locator #{tok} ignored"
119 elsif Blob.new(tok).empty?
120 # No signature provided -- but no data to protect, either.
122 logger.warn "Missing signature on locator #{tok}"
123 raise ArvadosModel::PermissionDeniedError
128 @signatures_checked = computed_pdh
131 def strip_signatures_and_update_replication_confirmed
132 if self.manifest_text_changed?
134 if not self.replication_confirmed.nil?
135 self.class.each_manifest_locator(manifest_text_was) do |match|
136 in_old_manifest[match[1]] = true
140 stripped_manifest = self.class.munge_manifest_locators(manifest_text) do |match|
141 if not self.replication_confirmed.nil? and not in_old_manifest[match[1]]
142 # If the new manifest_text contains locators whose hashes
143 # weren't in the old manifest_text, storage replication is no
145 self.replication_confirmed_at = nil
146 self.replication_confirmed = nil
149 # Return the locator with all permission signatures removed,
150 # but otherwise intact.
151 match[0].gsub(/\+A[^+]*/, '')
154 if @computed_pdh_for_manifest_text == manifest_text
155 # If the cached PDH was valid before stripping, it is still
156 # valid after stripping.
157 @computed_pdh_for_manifest_text = stripped_manifest.dup
160 self[:manifest_text] = stripped_manifest
165 def ensure_pdh_matches_manifest_text
166 if not manifest_text_changed? and not portable_data_hash_changed?
168 elsif portable_data_hash.nil? or not portable_data_hash_changed?
169 self.portable_data_hash = computed_pdh
170 elsif portable_data_hash !~ Keep::Locator::LOCATOR_REGEXP
171 errors.add(:portable_data_hash, "is not a valid locator")
173 elsif portable_data_hash[0..31] != computed_pdh[0..31]
174 errors.add(:portable_data_hash,
175 "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'")
178 # Ignore the client-provided size part: always store
179 # computed_pdh in the database.
180 self.portable_data_hash = computed_pdh
185 if self.manifest_text_changed?
186 self.file_names = manifest_files
193 if self.manifest_text
194 self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
195 names << name.first.gsub('\040',' ') + "\n"
196 break if names.length > 2**12
200 if self.manifest_text and names.length < 2**12
201 self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
202 names << stream_name.first.gsub('\040',' ') + "\n"
203 break if names.length > 2**12
210 def default_empty_manifest
211 self.manifest_text ||= ''
215 if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
219 # If Ruby thinks the encoding is something else, like 7-bit
220 # ASCII, but its stored bytes are equal to the (valid) UTF-8
221 # encoding of the same string, we declare it to be a UTF-8
224 utf8.force_encoding Encoding::UTF_8
225 if utf8.valid_encoding? and utf8 == manifest_text.encode(Encoding::UTF_8)
226 self.manifest_text = utf8
231 errors.add :manifest_text, "must use UTF-8 encoding"
236 def check_manifest_validity
238 Keep::Manifest.validate! manifest_text
240 rescue ArgumentError => e
241 errors.add :manifest_text, e.message
246 def signed_manifest_text
247 if !has_attribute? :manifest_text
252 token = current_api_client_authorization.andand.api_token
253 exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
254 trash_at].compact.map(&:to_i).min
255 self.class.sign_manifest manifest_text, token, exp
259 def self.sign_manifest manifest, token, exp=nil
261 exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl
267 m = munge_manifest_locators(manifest) do |match|
268 Blob.sign_locator(match[0], signing_opts)
273 def self.munge_manifest_locators manifest
274 # Given a manifest text and a block, yield the regexp MatchData
275 # for each locator. Return a new manifest in which each locator
276 # has been replaced by the block's return value.
277 return nil if !manifest
278 return '' if manifest == ''
281 manifest.each_line do |line|
284 line.split(' ').each do |word|
287 elsif match = Keep::Locator::LOCATOR_REGEXP.match(word)
288 new_words << yield(match)
293 new_lines << new_words.join(' ')
295 new_lines.join("\n") + "\n"
298 def self.each_manifest_locator manifest
299 # Given a manifest text and a block, yield the regexp match object
301 manifest.each_line do |line|
302 # line will have a trailing newline, but the last token is never
303 # a locator, so it's harmless here.
304 line.split(' ').each do |word|
305 if match = Keep::Locator::LOCATOR_REGEXP.match(word)
312 def self.normalize_uuid uuid
315 uuid.split('+').each do |token|
316 if token.match(/^[0-9a-f]{32,}$/)
317 raise "uuid #{uuid} has multiple hash parts" if hash_part
319 elsif token.match(/^\d+$/)
320 raise "uuid #{uuid} has multiple size parts" if size_part
324 raise "uuid #{uuid} has no hash part" if !hash_part
325 [hash_part, size_part].compact.join '+'
328 def self.get_compatible_images(readers, pattern, collections)
329 if collections.empty?
334 Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?',
335 collections.map(&:portable_data_hash),
336 'docker_image_migration',
338 order('links.created_at asc').
340 [l.tail_uuid, l.head_uuid]
343 migrated_collections = Hash[
344 Collection.readable_by(*readers).
345 where('portable_data_hash in (?)', migrations.values).
347 [c.portable_data_hash, c]
350 collections.map { |c|
351 # Check if the listed image is compatible first, if not, then try the
353 manifest = Keep::Manifest.new(c.manifest_text)
354 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
356 elsif m = migrated_collections[migrations[c.portable_data_hash]]
357 manifest = Keep::Manifest.new(m.manifest_text)
358 if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
365 # Resolve a Docker repo+tag, hash, or collection PDH to an array of
366 # Collection objects, sorted by timestamp starting with the most recent
369 # If filter_compatible_format is true (the default), only return image
370 # collections which are support by the installation as indicated by
371 # Rails.configuration.docker_image_formats. Will follow
372 # 'docker_image_migration' links if search_term resolves to an incompatible
373 # image, but an equivalent compatible image is available.
374 def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
375 readers ||= [Thread.current[:user]]
377 readable_by(*readers).
378 readable_by(*readers, table_name: "collections").
379 joins("JOIN collections ON links.head_uuid = collections.uuid").
380 order("links.created_at DESC")
382 if (Rails.configuration.docker_image_formats.include? 'v1' and
383 Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false
384 pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
385 elsif Rails.configuration.docker_image_formats.include? 'v2'
386 pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
387 elsif Rails.configuration.docker_image_formats.include? 'v1'
388 pattern = /^[0-9A-Fa-f]{64}\.tar$/
390 raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}"
393 # If the search term is a Collection locator that contains one file
394 # that looks like a Docker image, return it.
395 if loc = Keep::Locator.parse(search_term)
397 coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
398 return get_compatible_images(readers, pattern, coll_match)
401 if search_tag.nil? and (n = search_term.index(":"))
402 search_tag = search_term[n+1..-1]
403 search_term = search_term[0..n-1]
406 # Find Collections with matching Docker image repository+tag pairs.
407 matches = base_search.
408 where(link_class: "docker_image_repo+tag",
409 name: "#{search_term}:#{search_tag || 'latest'}")
411 # If that didn't work, find Collections with matching Docker image hashes.
413 matches = base_search.
414 where("link_class = ? and links.name LIKE ?",
415 "docker_image_hash", "#{search_term}%")
418 # Generate an order key for each result. We want to order the results
419 # so that anything with an image timestamp is considered more recent than
420 # anything without; then we use the link's created_at as a tiebreaker.
422 matches.each do |link|
423 uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
424 -link.created_at.to_i]
427 sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c|
428 uuid_timestamps[c.uuid]
430 compatible = get_compatible_images(readers, pattern, sorted)
431 if sorted.length > 0 and compatible.empty?
432 raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration."
437 def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
438 find_all_for_docker_image(search_term, search_tag, readers).first
441 def self.searchable_columns operator
442 super - ["manifest_text"]
445 def self.full_text_searchable_columns
446 super - ["manifest_text", "storage_classes_desired", "storage_classes_confirmed"]
450 SweepTrashedCollections.sweep_if_stale
456 def portable_manifest_text
457 self.class.munge_manifest_locators(manifest_text) do |match|
467 portable_manifest = portable_manifest_text
468 (Digest::MD5.hexdigest(portable_manifest) +
470 portable_manifest.bytesize.to_s)
474 if @computed_pdh_for_manifest_text == manifest_text
477 @computed_pdh = compute_pdh
478 @computed_pdh_for_manifest_text = manifest_text.dup
482 def ensure_permission_to_save
483 if (not current_user.andand.is_admin)
484 if (replication_confirmed_at_changed? or replication_confirmed_changed?) and
485 not (replication_confirmed_at.nil? and replication_confirmed.nil?)
486 raise ArvadosModel::PermissionDeniedError.new("replication_confirmed and replication_confirmed_at attributes cannot be changed, except by setting both to nil")
488 if (storage_classes_confirmed_changed? or storage_classes_confirmed_at_changed?) and
489 not (storage_classes_confirmed == [] and storage_classes_confirmed_at.nil?)
490 raise ArvadosModel::PermissionDeniedError.new("storage_classes_confirmed and storage_classes_confirmed_at attributes cannot be changed, except by setting them to [] and nil respectively")
496 def ensure_storage_classes_desired_is_not_empty
497 if self.storage_classes_desired.empty?
498 raise ArvadosModel::InvalidStateTransitionError.new("storage_classes_desired shouldn't be empty")
502 def ensure_storage_classes_contain_non_empty_strings
503 (self.storage_classes_desired + self.storage_classes_confirmed).each do |c|
504 if !c.is_a?(String) || c == ''
505 raise ArvadosModel::InvalidStateTransitionError.new("storage classes should only be non-empty strings")