Merge branch '8784-dir-listings'
[arvados.git] / services / api / app / models / collection.rb
index 4f774d6f9cc7029759a474938da991cbc83c2f46..6f13a6365127191020c375679ccc985b7d85db4e 100644 (file)
@@ -1,7 +1,12 @@
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
 require 'arvados/keep'
 require 'sweep_trashed_collections'
 
 class Collection < ArvadosModel
+  extend CurrentApiClient
   extend DbCurrentTime
   include HasUuid
   include KindAndEtag
@@ -23,7 +28,7 @@ class Collection < ArvadosModel
   before_save :set_file_names
 
   # Query only untrashed collections by default.
-  default_scope where("is_trashed = false")
+  default_scope { where("is_trashed = false") }
 
   api_accessible :user, extend: :common do |t|
     t.add :name
@@ -31,6 +36,7 @@ class Collection < ArvadosModel
     t.add :properties
     t.add :portable_data_hash
     t.add :signed_manifest_text, as: :manifest_text
+    t.add :manifest_text, as: :unsigned_manifest_text
     t.add :replication_desired
     t.add :replication_confirmed
     t.add :replication_confirmed_at
@@ -55,6 +61,7 @@ class Collection < ArvadosModel
                 # We need trash_at and is_trashed to determine the
                 # correct timestamp in signed_manifest_text.
                 'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
+                'unsigned_manifest_text' => ['manifest_text'],
                 )
   end
 
@@ -62,6 +69,10 @@ class Collection < ArvadosModel
     super + ["updated_at", "file_names"]
   end
 
+  def self.limit_index_columns_read
+    ["manifest_text"]
+  end
+
   FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
   def check_signatures
     return false if self.manifest_text.nil?
@@ -160,7 +171,7 @@ class Collection < ArvadosModel
       false
     elsif portable_data_hash[0..31] != computed_pdh[0..31]
       errors.add(:portable_data_hash,
-                 "does not match computed hash #{computed_pdh}")
+                 "'#{portable_data_hash}' does not match computed hash '#{computed_pdh}'")
       false
     else
       # Ignore the client-provided size part: always store
@@ -313,8 +324,53 @@ class Collection < ArvadosModel
     [hash_part, size_part].compact.join '+'
   end
 
-  # Return array of Collection objects
-  def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil)
+  def self.get_compatible_images(readers, pattern, collections)
+    if collections.empty?
+      return []
+    end
+
+    migrations = Hash[
+      Link.where('tail_uuid in (?) AND link_class=? AND links.owner_uuid=?',
+                 collections.map(&:portable_data_hash),
+                 'docker_image_migration',
+                 system_user_uuid).
+      order('links.created_at asc').
+      map { |l|
+        [l.tail_uuid, l.head_uuid]
+      }]
+
+    migrated_collections = Hash[
+      Collection.readable_by(*readers).
+      where('portable_data_hash in (?)', migrations.values).
+      map { |c|
+        [c.portable_data_hash, c]
+      }]
+
+    collections.map { |c|
+      # Check if the listed image is compatible first, if not, then try the
+      # migration link.
+      manifest = Keep::Manifest.new(c.manifest_text)
+      if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
+        c
+      elsif m = migrated_collections[migrations[c.portable_data_hash]]
+        manifest = Keep::Manifest.new(m.manifest_text)
+        if manifest.exact_file_count?(1) and manifest.files[0][1] =~ pattern
+          m
+        end
+      end
+    }.compact
+  end
+
+  # Resolve a Docker repo+tag, hash, or collection PDH to an array of
+  # Collection objects, sorted by timestamp starting with the most recent
+  # match.
+  #
+  # If filter_compatible_format is true (the default), only return image
+  # collections which are support by the installation as indicated by
+  # Rails.configuration.docker_image_formats.  Will follow
+  # 'docker_image_migration' links if search_term resolves to an incompatible
+  # image, but an equivalent compatible image is available.
+  def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
     readers ||= [Thread.current[:user]]
     base_search = Link.
       readable_by(*readers).
@@ -322,20 +378,23 @@ class Collection < ArvadosModel
       joins("JOIN collections ON links.head_uuid = collections.uuid").
       order("links.created_at DESC")
 
+    if (Rails.configuration.docker_image_formats.include? 'v1' and
+        Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false
+      pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
+    elsif Rails.configuration.docker_image_formats.include? 'v2'
+      pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
+    elsif Rails.configuration.docker_image_formats.include? 'v1'
+      pattern = /^[0-9A-Fa-f]{64}\.tar$/
+    else
+      raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}"
+    end
+
     # If the search term is a Collection locator that contains one file
     # that looks like a Docker image, return it.
     if loc = Keep::Locator.parse(search_term)
       loc.strip_hints!
-      coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1).first
-      if coll_match
-        # Check if the Collection contains exactly one file whose name
-        # looks like a saved Docker image.
-        manifest = Keep::Manifest.new(coll_match.manifest_text)
-        if manifest.exact_file_count?(1) and
-            (manifest.files[0][1] =~ /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/)
-          return [coll_match]
-        end
-      end
+      coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
+      return get_compatible_images(readers, pattern, coll_match)
     end
 
     if search_tag.nil? and (n = search_term.index(":"))
@@ -359,11 +418,19 @@ class Collection < ArvadosModel
     # so that anything with an image timestamp is considered more recent than
     # anything without; then we use the link's created_at as a tiebreaker.
     uuid_timestamps = {}
-    matches.all.map do |link|
+    matches.each do |link|
       uuid_timestamps[link.head_uuid] = [(-link.properties["image_timestamp"].to_datetime.to_i rescue 0),
        -link.created_at.to_i]
+     end
+
+    sorted = Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c|
+      uuid_timestamps[c.uuid]
+    }
+    compatible = get_compatible_images(readers, pattern, sorted)
+    if sorted.length > 0 and compatible.empty?
+      raise ArvadosModel::UnresolvableContainerError.new "Matching Docker image is incompatible with 'docker_image_formats' configuration."
     end
-    Collection.where('uuid in (?)', uuid_timestamps.keys).sort_by { |c| uuid_timestamps[c.uuid] }
+    compatible
   end
 
   def self.for_latest_docker_image(search_term, search_tag=nil, readers=nil)
@@ -451,33 +518,49 @@ class Collection < ArvadosModel
     true
   end
 
-  # If trash_at is updated without touching delete_at, automatically
-  # update delete_at to a sensible value.
   def default_trash_interval
     if trash_at_changed? && !delete_at_changed?
+      # If trash_at is updated without touching delete_at,
+      # automatically update delete_at to a sensible value.
       if trash_at.nil?
         self.delete_at = nil
       else
         self.delete_at = trash_at + Rails.configuration.default_trash_lifetime.seconds
       end
+    elsif !trash_at || !delete_at || trash_at > delete_at
+      # Not trash, or bogus arguments? Just validate in
+      # validate_trash_and_delete_timing.
+    elsif delete_at_changed? && delete_at >= trash_at
+      # Fix delete_at if needed, so it's not earlier than the expiry
+      # time on any permission tokens that might have been given out.
+
+      # In any case there are no signatures expiring after now+TTL.
+      # Also, if the existing trash_at time has already passed, we
+      # know we haven't given out any signatures since then.
+      earliest_delete = [
+        @validation_timestamp,
+        trash_at_was,
+      ].compact.min + Rails.configuration.blob_signature_ttl.seconds
+
+      # The previous value of delete_at is also an upper bound on the
+      # longest-lived permission token. For example, if TTL=14,
+      # trash_at_was=now-7, delete_at_was=now+7, then it is safe to
+      # set trash_at=now+6, delete_at=now+8.
+      earliest_delete = [earliest_delete, delete_at_was].compact.min
+
+      # If delete_at is too soon, use the earliest possible time.
+      if delete_at < earliest_delete
+        self.delete_at = earliest_delete
+      end
     end
   end
 
   def validate_trash_and_delete_timing
     if trash_at.nil? != delete_at.nil?
       errors.add :delete_at, "must be set if trash_at is set, and must be nil otherwise"
-    end
-
-    earliest_delete = ([@validation_timestamp, trash_at_was].compact.min +
-                       Rails.configuration.blob_signature_ttl.seconds)
-    if delete_at && delete_at < earliest_delete
-      errors.add :delete_at, "#{delete_at} is too soon: earliest allowed is #{earliest_delete}"
-    end
-
-    if delete_at && delete_at < trash_at
+    elsif delete_at && delete_at < trash_at
       errors.add :delete_at, "must not be earlier than trash_at"
     end
-
     true
   end
 end