<meta name="author" content="">
<% if current_user %>
<% content_for :js do %>
- window.defaultSession = <%=raw({baseURL: Rails.configuration.Services.Controller.ExternalURL.to_s.gsub(/\/?$/,'/'), token: Thread.current[:arvados_api_token], user: current_user}.to_json)%>
+ window.defaultSession = <%=raw({baseURL: Rails.configuration.Services.Controller.ExternalURL.to_s.sub(/\/*$/,'/'), token: Thread.current[:arvados_api_token], user: current_user}.to_json)%>
<% end %>
<% end %>
<% if current_user and $arvados_api_client.discovery[:websocketUrl] %>
<%= yield :head %>
<%= javascript_tag do %>
angular.module('Arvados').value('arvadosApiToken', '<%=Thread.current[:arvados_api_token]%>');
- angular.module('Arvados').value('arvadosDiscoveryUri', '<%= Rails.configuration.Services.Controller.ExternalURL.to_s + '/discovery/v1/apis/arvados/v1/rest' %>');
+ angular.module('Arvados').value('arvadosDiscoveryUri', '<%= Rails.configuration.Services.Controller.ExternalURL.to_s.sub(/\/*$/,'/') + 'discovery/v1/apis/arvados/v1/rest' %>');
<%= yield :js %>
<% end %>
<style>
--upload
If the build and test steps are successful, upload the python
packages to pypi and the gems to rubygems (default: false)
+--ruby <true|false>
+ Build ruby gems (default: true)
+--python <true|false>
+ Build python packages (default: true)
WORKSPACE=path Path to the Arvados source tree to build packages from
TARGET=
UPLOAD=0
+RUBY=1
+PYTHON=1
DEBUG=${ARVADOS_DEBUG:-0}
PARSEDOPTS=$(getopt --name "$0" --longoptions \
- help,debug,upload,target: \
+ help,debug,ruby:,python:,upload,target: \
-- "" "$@")
if [ $? -ne 0 ]; then
exit 1
--target)
TARGET="$2"; shift
;;
+ --ruby)
+ RUBY="$2"; shift
+ if [ "$RUBY" != "true" ] && [ "$RUBY" != "1" ]; then
+ RUBY=0
+ else
+ RUBY=1
+ fi
+ ;;
+ --python)
+ PYTHON="$2"; shift
+ if [ "$PYTHON" != "true" ] && [ "$PYTHON" != "1" ]; then
+ PYTHON=0
+ else
+ PYTHON=1
+ fi
+ ;;
--upload)
UPLOAD=1
;;
debug_echo "$0 is running from $RUN_BUILD_PACKAGES_PATH"
debug_echo "Workspace is $WORKSPACE"
+if [ $RUBY -eq 0 ] && [ $PYTHON -eq 0 ]; then
+ echo "Nothing to do!"
+ exit 0
+fi
+
if [[ -f /etc/profile.d/rvm.sh ]]; then
source /etc/profile.d/rvm.sh
GEM="rvm-exec default gem"
debug_echo "umask is" `umask`
-gem_wrapper arvados "$WORKSPACE/sdk/ruby"
-gem_wrapper arvados-cli "$WORKSPACE/sdk/cli"
-gem_wrapper arvados-login-sync "$WORKSPACE/services/login-sync"
-
GEM_BUILD_FAILURES=0
-if [ ${#failures[@]} -ne 0 ]; then
- GEM_BUILD_FAILURES=${#failures[@]}
+if [ $RUBY -eq 1 ]; then
+ debug_echo "Building Ruby gems"
+ gem_wrapper arvados "$WORKSPACE/sdk/ruby"
+ gem_wrapper arvados-cli "$WORKSPACE/sdk/cli"
+ gem_wrapper arvados-login-sync "$WORKSPACE/services/login-sync"
+ if [ ${#failures[@]} -ne 0 ]; then
+ GEM_BUILD_FAILURES=${#failures[@]}
+ fi
fi
-python_wrapper arvados-python-client "$WORKSPACE/sdk/python"
-python_wrapper arvados-pam "$WORKSPACE/sdk/pam"
-python_wrapper arvados-cwl-runner "$WORKSPACE/sdk/cwl"
-python_wrapper arvados_fuse "$WORKSPACE/services/fuse"
-python_wrapper arvados-node-manager "$WORKSPACE/services/nodemanager"
-
PYTHON_BUILD_FAILURES=0
-if [ $((${#failures[@]} - $GEM_BUILD_FAILURES)) -ne 0 ]; then
- PYTHON_BUILD_FAILURES=${#failures[@]} - $GEM_BUILD_FAILURES
+if [ $PYTHON -eq 1 ]; then
+ debug_echo "Building Python packages"
+ python_wrapper arvados-python-client "$WORKSPACE/sdk/python"
+ python_wrapper arvados-pam "$WORKSPACE/sdk/pam"
+ python_wrapper arvados-cwl-runner "$WORKSPACE/sdk/cwl"
+ python_wrapper arvados_fuse "$WORKSPACE/services/fuse"
+ python_wrapper arvados-node-manager "$WORKSPACE/services/nodemanager"
+
+ if [ $((${#failures[@]} - $GEM_BUILD_FAILURES)) -ne 0 ]; then
+ PYTHON_BUILD_FAILURES=$((${#failures[@]} - $GEM_BUILD_FAILURES))
+ fi
fi
-if [[ "$UPLOAD" != 0 ]]; then
+if [ $UPLOAD -ne 0 ]; then
+ echo "Uploading"
- if [[ $DEBUG > 0 ]]; then
+ if [ $DEBUG > 0 ]; then
EXTRA_UPLOAD_FLAGS=" --verbose"
else
EXTRA_UPLOAD_FLAGS=""
fi
- if [[ ! -e "$WORKSPACE/packages" ]]; then
+ if [ ! -e "$WORKSPACE/packages" ]; then
mkdir -p "$WORKSPACE/packages"
fi
- title "Start upload python packages"
- timer_reset
-
- if [ "$PYTHON_BUILD_FAILURES" -eq 0 ]; then
- /usr/local/arvados-dev/jenkins/run_upload_packages.py $EXTRA_UPLOAD_FLAGS --workspace $WORKSPACE python
- else
- echo "Skipping python packages upload, there were errors building the packages"
+ if [ $PYTHON -eq 1 ]; then
+ title "Start upload python packages"
+ timer_reset
+
+ if [ $PYTHON_BUILD_FAILURES -eq 0 ]; then
+ /usr/local/arvados-dev/jenkins/run_upload_packages.py $EXTRA_UPLOAD_FLAGS --workspace $WORKSPACE python
+ else
+ echo "Skipping python packages upload, there were errors building the packages"
+ fi
+ checkexit $? "upload python packages"
+ title "End of upload python packages (`timer`)"
fi
- checkexit $? "upload python packages"
- title "End of upload python packages (`timer`)"
- title "Start upload ruby gems"
- timer_reset
-
- if [ "$GEM_BUILD_FAILURES" -eq 0 ]; then
- /usr/local/arvados-dev/jenkins/run_upload_packages.py $EXTRA_UPLOAD_FLAGS --workspace $WORKSPACE gems
- else
- echo "Skipping ruby gem upload, there were errors building the packages"
+ if [ $RUBY -eq 1 ]; then
+ title "Start upload ruby gems"
+ timer_reset
+
+ if [ $GEM_BUILD_FAILURES -eq 0 ]; then
+ /usr/local/arvados-dev/jenkins/run_upload_packages.py $EXTRA_UPLOAD_FLAGS --workspace $WORKSPACE gems
+ else
+ echo "Skipping ruby gem upload, there were errors building the packages"
+ fi
+ checkexit $? "upload ruby gems"
+ title "End of upload ruby gems (`timer`)"
fi
- checkexit $? "upload ruby gems"
- title "End of upload ruby gems (`timer`)"
-
fi
exit_cleanly
"git.arvados.org/arvados.git/lib/crunchrun"
"git.arvados.org/arvados.git/lib/dispatchcloud"
"git.arvados.org/arvados.git/lib/install"
- "git.arvados.org/arvados.git/lib/undelete"
+ "git.arvados.org/arvados.git/lib/recovercollection"
"git.arvados.org/arvados.git/services/ws"
)
"-version": cmd.Version,
"--version": cmd.Version,
- "boot": boot.Command,
- "cloudtest": cloudtest.Command,
- "config-check": config.CheckCommand,
- "config-defaults": config.DumpDefaultsCommand,
- "config-dump": config.DumpCommand,
- "controller": controller.Command,
- "crunch-run": crunchrun.Command,
- "dispatch-cloud": dispatchcloud.Command,
- "install": install.Command,
- "undelete": undelete.Command,
- "ws": ws.Command,
+ "boot": boot.Command,
+ "cloudtest": cloudtest.Command,
+ "config-check": config.CheckCommand,
+ "config-defaults": config.DumpDefaultsCommand,
+ "config-dump": config.DumpCommand,
+ "controller": controller.Command,
+ "crunch-run": crunchrun.Command,
+ "dispatch-cloud": dispatchcloud.Command,
+ "install": install.Command,
+ "recover-collection": recovercollection.Command,
+ "ws": ws.Command,
})
)
- admin/logs-table-management.html.textile.liquid
- admin/workbench2-vocabulary.html.textile.liquid
- admin/storage-classes.html.textile.liquid
+ - admin/recovering-deleted-collections.html.textile.liquid
- Cloud:
- admin/spot-instances.html.textile.liquid
- admin/cloudtest.html.textile.liquid
--- /dev/null
+---
+layout: default
+navsection: admin
+title: Recovering deleted collections
+...
+
+{% comment %}
+Copyright (C) The Arvados Authors. All rights reserved.
+
+SPDX-License-Identifier: CC-BY-SA-3.0
+{% endcomment %}
+
+In some cases, it is possible to recover files that have been lost by modifying or deleting a collection.
+
+Possibility of recovery depends on many factors, including:
+* Whether the collection manifest is still available, e.g., in an audit log entry
+* Whether the data blocks are also referenced by other collections
+* Whether the data blocks have been unreferenced long enough to be marked for deletion/trash by keep-balance
+* Blob signature TTL, trash lifetime, trash check interval, and other config settings
+
+To attempt recovery of a previous version of a deleted/modified collection, use the @arvados-server recover-collection@ command. It should be run on one of your server nodes where the @arvados-server@ package is installed and the @/etc/arvados/config.yml@ file is up to date.
+
+Specify the collection you want to recover by passing either the UUID of an audit log entry, or a file containing the manifest.
+
+If recovery is successful, the @recover-collection@ program saves the recovered data a new collection belonging to the system user, and prints the new collection's UUID on stdout.
+
+<pre>
+# arvados-server recover-collection 9tee4-57u5n-nb5awmk1pahac2t
+INFO[2020-06-05T19:52:29.557761245Z] loaded log entry logged_event_time="2020-06-05 16:48:01.438791 +0000 UTC" logged_event_type=update old_collection_uuid=9tee4-4zz18-1ex26g95epmgw5w src=9tee4-57u5n-nb5awmk1pahac2t
+INFO[2020-06-05T19:52:29.642145127Z] recovery succeeded UUID=9tee4-4zz18-5trfp4k4xxg97f1 src=9tee4-57u5n-nb5awmk1pahac2t
+9tee4-4zz18-5trfp4k4xxg97f1
+INFO[2020-06-05T19:52:29.644699436Z] exiting
+</pre>
+
+In this example, the original data has been restored and saved in a new collection with UUID @9tee4-4zz18-5trfp4k4xxg97f1@.
+
+For more options, run @arvados-server recover-collection -help@.
ClientSecret: "zzzzzzzzzzzzzzzzzzzzzzzz"
</pre>
+Check the OpenIDConnect section in the "default config file":{{site.baseurl}}/admin/config.html for more details and configuration options.
+
h2(#ldap). LDAP
With this configuration, authentication uses an external LDAP service like OpenLDAP or Active Directory.
ClientID: ""
ClientSecret: ""
+ # OpenID claim field containing the user's email
+ # address. Normally "email"; see
+ # https://openid.net/specs/openid-connect-core-1_0.html#StandardClaims
+ EmailClaim: "email"
+
+ # OpenID claim field containing the email verification
+ # flag. Normally "email_verified". To accept every returned
+ # email address without checking a "verified" field at all,
+ # use the empty string "".
+ EmailVerifiedClaim: "email_verified"
+
+ # OpenID claim field containing the user's preferred
+ # username. If empty, use the mailbox part of the user's email
+ # address.
+ UsernameClaim: ""
+
PAM:
# (Experimental) Use PAM to authenticate users.
Enable: false
"Login.OpenIDConnect.ClientSecret": false,
"Login.OpenIDConnect.Enable": true,
"Login.OpenIDConnect.Issuer": false,
+ "Login.OpenIDConnect.EmailClaim": false,
+ "Login.OpenIDConnect.EmailVerifiedClaim": false,
+ "Login.OpenIDConnect.UsernameClaim": false,
"Login.PAM": true,
"Login.PAM.DefaultEmailDomain": false,
"Login.PAM.Enable": true,
ClientID: ""
ClientSecret: ""
+ # OpenID claim field containing the user's email
+ # address. Normally "email"; see
+ # https://openid.net/specs/openid-connect-core-1_0.html#StandardClaims
+ EmailClaim: "email"
+
+ # OpenID claim field containing the email verification
+ # flag. Normally "email_verified". To accept every returned
+ # email address without checking a "verified" field at all,
+ # use the empty string "".
+ EmailVerifiedClaim: "email_verified"
+
+ # OpenID claim field containing the user's preferred
+ # username. If empty, use the mailbox part of the user's email
+ # address.
+ UsernameClaim: ""
+
PAM:
# (Experimental) Use PAM to authenticate users.
Enable: false
ClientID: cluster.Login.Google.ClientID,
ClientSecret: cluster.Login.Google.ClientSecret,
UseGooglePeopleAPI: cluster.Login.Google.AlternateEmailAddresses,
+ EmailClaim: "email",
+ EmailVerifiedClaim: "email_verified",
}
case !wantGoogle && wantOpenIDConnect && !wantSSO && !wantPAM && !wantLDAP:
return &oidcLoginController{
- Cluster: cluster,
- RailsProxy: railsProxy,
- Issuer: cluster.Login.OpenIDConnect.Issuer,
- ClientID: cluster.Login.OpenIDConnect.ClientID,
- ClientSecret: cluster.Login.OpenIDConnect.ClientSecret,
+ Cluster: cluster,
+ RailsProxy: railsProxy,
+ Issuer: cluster.Login.OpenIDConnect.Issuer,
+ ClientID: cluster.Login.OpenIDConnect.ClientID,
+ ClientSecret: cluster.Login.OpenIDConnect.ClientSecret,
+ EmailClaim: cluster.Login.OpenIDConnect.EmailClaim,
+ EmailVerifiedClaim: cluster.Login.OpenIDConnect.EmailVerifiedClaim,
+ UsernameClaim: cluster.Login.OpenIDConnect.UsernameClaim,
}
case !wantGoogle && !wantOpenIDConnect && wantSSO && !wantPAM && !wantLDAP:
return &ssoLoginController{railsProxy}
Issuer string // OIDC issuer URL, e.g., "https://accounts.google.com"
ClientID string
ClientSecret string
- UseGooglePeopleAPI bool // Use Google People API to look up alternate email addresses
+ UseGooglePeopleAPI bool // Use Google People API to look up alternate email addresses
+ EmailClaim string // OpenID claim to use as email address; typically "email"
+ EmailVerifiedClaim string // If non-empty, ensure claim value is true before accepting EmailClaim; typically "email_verified"
+ UsernameClaim string // If non-empty, use as preferred username
// override Google People API base URL for testing purposes
// (normally empty, set by google pkg to
var ret rpc.UserSessionAuthInfo
defer ctxlog.FromContext(ctx).WithField("ret", &ret).Debug("getAuthInfo returned")
- var claims struct {
- Name string `json:"name"`
- Email string `json:"email"`
- Verified bool `json:"email_verified"`
- }
+ var claims map[string]interface{}
if err := idToken.Claims(&claims); err != nil {
return nil, fmt.Errorf("error extracting claims from ID token: %s", err)
- } else if claims.Verified {
+ } else if verified, _ := claims[ctrl.EmailVerifiedClaim].(bool); verified || ctrl.EmailVerifiedClaim == "" {
// Fall back to this info if the People API call
// (below) doesn't return a primary && verified email.
- if names := strings.Fields(strings.TrimSpace(claims.Name)); len(names) > 1 {
+ name, _ := claims["name"].(string)
+ if names := strings.Fields(strings.TrimSpace(name)); len(names) > 1 {
ret.FirstName = strings.Join(names[0:len(names)-1], " ")
ret.LastName = names[len(names)-1]
} else {
ret.FirstName = names[0]
}
- ret.Email = claims.Email
+ ret.Email, _ = claims[ctrl.EmailClaim].(string)
+ }
+
+ if ctrl.UsernameClaim != "" {
+ ret.Username, _ = claims[ctrl.UsernameClaim].(string)
}
if !ctrl.UseGooglePeopleAPI {
if ret.Email == "" {
- return nil, fmt.Errorf("cannot log in with unverified email address %q", claims.Email)
+ return nil, fmt.Errorf("cannot log in with unverified email address %q", claims[ctrl.EmailClaim])
}
return &ret, nil
}
return nil, errors.New("cannot log in without a verified email address")
}
for ae := range altEmails {
- if ae != ret.Email {
- ret.AlternateEmails = append(ret.AlternateEmails, ae)
- if i := strings.Index(ae, "@"); i > 0 && strings.ToLower(ae[i+1:]) == strings.ToLower(ctrl.Cluster.Users.PreferDomainForUsername) {
+ if ae == ret.Email {
+ continue
+ }
+ ret.AlternateEmails = append(ret.AlternateEmails, ae)
+ if ret.Username == "" {
+ i := strings.Index(ae, "@")
+ if i > 0 && strings.ToLower(ae[i+1:]) == strings.ToLower(ctrl.Cluster.Users.PreferDomainForUsername) {
ret.Username = strings.SplitN(ae[:i], "+", 2)[0]
}
}
"email": s.authEmail,
"email_verified": s.authEmailVerified,
"name": s.authName,
+ "alt_verified": true, // for custom claim tests
+ "alt_email": "alt_email@example.com", // for custom claim tests
+ "alt_username": "desired-username", // for custom claim tests
})
json.NewEncoder(w).Encode(struct {
AccessToken string `json:"access_token"`
c.Check(resp.RedirectLocation, check.Equals, "")
}
-func (s *OIDCLoginSuite) TestOIDCLogin_Success(c *check.C) {
+func (s *OIDCLoginSuite) TestGenericOIDCLogin(c *check.C) {
s.cluster.Login.Google.Enable = false
s.cluster.Login.OpenIDConnect.Enable = true
json.Unmarshal([]byte(fmt.Sprintf("%q", s.fakeIssuer.URL)), &s.cluster.Login.OpenIDConnect.Issuer)
s.cluster.Login.OpenIDConnect.ClientSecret = "oidc#client#secret"
s.validClientID = "oidc#client#id"
s.validClientSecret = "oidc#client#secret"
- s.localdb = NewConn(s.cluster)
- state := s.startLogin(c)
- resp, err := s.localdb.Login(context.Background(), arvados.LoginOptions{
- Code: s.validCode,
- State: state,
- })
- c.Assert(err, check.IsNil)
- c.Check(resp.HTML.String(), check.Equals, "")
- target, err := url.Parse(resp.RedirectLocation)
- c.Assert(err, check.IsNil)
- token := target.Query().Get("api_token")
- c.Check(token, check.Matches, `v2/zzzzz-gj3su-.{15}/.{32,50}`)
+ for _, trial := range []struct {
+ expectEmail string // "" if failure expected
+ setup func()
+ }{
+ {
+ expectEmail: "user@oidc.example.com",
+ setup: func() {
+ c.Log("=== succeed because email_verified is false but not required")
+ s.authEmail = "user@oidc.example.com"
+ s.authEmailVerified = false
+ s.cluster.Login.OpenIDConnect.EmailClaim = "email"
+ s.cluster.Login.OpenIDConnect.EmailVerifiedClaim = ""
+ s.cluster.Login.OpenIDConnect.UsernameClaim = ""
+ },
+ },
+ {
+ expectEmail: "",
+ setup: func() {
+ c.Log("=== fail because email_verified is false and required")
+ s.authEmail = "user@oidc.example.com"
+ s.authEmailVerified = false
+ s.cluster.Login.OpenIDConnect.EmailClaim = "email"
+ s.cluster.Login.OpenIDConnect.EmailVerifiedClaim = "email_verified"
+ s.cluster.Login.OpenIDConnect.UsernameClaim = ""
+ },
+ },
+ {
+ expectEmail: "user@oidc.example.com",
+ setup: func() {
+ c.Log("=== succeed because email_verified is false but config uses custom 'verified' claim")
+ s.authEmail = "user@oidc.example.com"
+ s.authEmailVerified = false
+ s.cluster.Login.OpenIDConnect.EmailClaim = "email"
+ s.cluster.Login.OpenIDConnect.EmailVerifiedClaim = "alt_verified"
+ s.cluster.Login.OpenIDConnect.UsernameClaim = ""
+ },
+ },
+ {
+ expectEmail: "alt_email@example.com",
+ setup: func() {
+ c.Log("=== succeed with custom 'email' and 'email_verified' claims")
+ s.authEmail = "bad@wrong.example.com"
+ s.authEmailVerified = false
+ s.cluster.Login.OpenIDConnect.EmailClaim = "alt_email"
+ s.cluster.Login.OpenIDConnect.EmailVerifiedClaim = "alt_verified"
+ s.cluster.Login.OpenIDConnect.UsernameClaim = "alt_username"
+ },
+ },
+ } {
+ trial.setup()
+ if s.railsSpy != nil {
+ s.railsSpy.Close()
+ }
+ s.railsSpy = arvadostest.NewProxy(c, s.cluster.Services.RailsAPI)
+ s.localdb = NewConn(s.cluster)
+ *s.localdb.railsProxy = *rpc.NewConn(s.cluster.ClusterID, s.railsSpy.URL, true, rpc.PassthroughTokenProvider)
+
+ state := s.startLogin(c)
+ resp, err := s.localdb.Login(context.Background(), arvados.LoginOptions{
+ Code: s.validCode,
+ State: state,
+ })
+ c.Assert(err, check.IsNil)
+ if trial.expectEmail == "" {
+ c.Check(resp.HTML.String(), check.Matches, `(?ms).*Login error.*`)
+ c.Check(resp.RedirectLocation, check.Equals, "")
+ continue
+ }
+ c.Check(resp.HTML.String(), check.Equals, "")
+ target, err := url.Parse(resp.RedirectLocation)
+ c.Assert(err, check.IsNil)
+ token := target.Query().Get("api_token")
+ c.Check(token, check.Matches, `v2/zzzzz-gj3su-.{15}/.{32,50}`)
+ authinfo := getCallbackAuthInfo(c, s.railsSpy)
+ c.Check(authinfo.Email, check.Equals, trial.expectEmail)
+
+ switch s.cluster.Login.OpenIDConnect.UsernameClaim {
+ case "alt_username":
+ c.Check(authinfo.Username, check.Equals, "desired-username")
+ case "":
+ c.Check(authinfo.Username, check.Equals, "")
+ default:
+ c.Fail() // bad test case
+ }
+ }
}
func (s *OIDCLoginSuite) TestGoogleLogin_Success(c *check.C) {
//
// SPDX-License-Identifier: AGPL-3.0
-package undelete
+package recovercollection
import (
"context"
flags.SetOutput(stderr)
flags.Usage = func() {
fmt.Fprintf(flags.Output(), `Usage:
- %s [options ...] /path/to/manifest.txt [...]
+ %s [options ...] { /path/to/manifest.txt | log-or-collection-uuid } [...]
This program recovers deleted collections. Recovery is
possible when the collection's manifest is still available and
collections, or the blocks have been trashed but not yet
deleted).
+ There are multiple ways to specify a collection to recover:
+
+ * Path to a local file containing a manifest with the desired
+ data
+
+ * UUID of an Arvados log entry, typically a "delete" or
+ "update" event, whose "old attributes" have a manifest with
+ the desired data
+
+ * UUID of an Arvados collection whose most recent log entry,
+ typically a "delete" or "update" event, has the desired
+ data in its "old attributes"
+
For each provided collection manifest, once all data blocks
are recovered/protected from garbage collection, a new
collection is saved and its UUID is printed on stdout.
+ Restored collections will belong to the system (root) user.
+
Exit status will be zero if recovery is successful, i.e., a
collection is saved for each provided manifest.
Options:
return 1
}
client.AuthToken = cluster.SystemRootToken
- und := undeleter{
+ rcvr := recoverer{
client: client,
cluster: cluster,
logger: logger,
exitcode := 0
for _, src := range flags.Args() {
logger := logger.WithField("src", src)
- if len(src) == 27 && src[5:12] == "-57u5n-" {
- logger.Error("log entry lookup not implemented")
- exitcode = 1
- continue
- } else {
- mtxt, err := ioutil.ReadFile(src)
+ var mtxt string
+ if !strings.Contains(src, "/") && len(src) == 27 && src[5] == '-' && src[11] == '-' {
+ var filters []arvados.Filter
+ if src[5:12] == "-57u5n-" {
+ filters = []arvados.Filter{{"uuid", "=", src}}
+ } else if src[5:12] == "-4zz18-" {
+ filters = []arvados.Filter{{"object_uuid", "=", src}}
+ } else {
+ logger.Error("looks like a UUID but not a log or collection UUID (if it's really a file, prepend './')")
+ exitcode = 1
+ continue
+ }
+ var resp struct {
+ Items []struct {
+ UUID string `json:"uuid"`
+ EventType string `json:"event_type"`
+ EventAt time.Time `json:"event_at"`
+ ObjectUUID string `json:"object_uuid"`
+ Properties struct {
+ OldAttributes struct {
+ ManifestText string `json:"manifest_text"`
+ } `json:"old_attributes"`
+ } `json:"properties"`
+ }
+ }
+ err = client.RequestAndDecode(&resp, "GET", "arvados/v1/logs", nil, arvados.ListOptions{
+ Limit: 1,
+ Order: []string{"event_at desc"},
+ Filters: filters,
+ })
if err != nil {
- logger.WithError(err).Error("error loading manifest data")
+ logger.WithError(err).Error("error looking up log entry")
+ exitcode = 1
+ continue
+ } else if len(resp.Items) == 0 {
+ logger.Error("log entry not found")
+ exitcode = 1
+ continue
+ }
+ logent := resp.Items[0]
+ logger.WithFields(logrus.Fields{
+ "uuid": logent.UUID,
+ "old_collection_uuid": logent.ObjectUUID,
+ "logged_event_type": logent.EventType,
+ "logged_event_time": logent.EventAt,
+ "logged_object_uuid": logent.ObjectUUID,
+ }).Info("loaded log entry")
+ mtxt = logent.Properties.OldAttributes.ManifestText
+ if mtxt == "" {
+ logger.Error("log entry properties.old_attributes.manifest_text missing or empty")
exitcode = 1
continue
}
- uuid, err := und.RecoverManifest(string(mtxt))
+ } else {
+ buf, err := ioutil.ReadFile(src)
if err != nil {
- logger.WithError(err).Error("recovery failed")
+ logger.WithError(err).Error("failed to load manifest data from file")
exitcode = 1
continue
}
- logger.WithField("UUID", uuid).Info("recovery succeeded")
- fmt.Fprintln(stdout, uuid)
+ mtxt = string(buf)
+ }
+ uuid, err := rcvr.RecoverManifest(string(mtxt))
+ if err != nil {
+ logger.WithError(err).Error("recovery failed")
+ exitcode = 1
+ continue
}
+ logger.WithField("UUID", uuid).Info("recovery succeeded")
+ fmt.Fprintln(stdout, uuid)
}
return exitcode
}
-type undeleter struct {
+type recoverer struct {
client *arvados.Client
cluster *arvados.Cluster
logger logrus.FieldLogger
// Finds the timestamp of the newest copy of blk on svc. Returns
// errNotFound if blk is not on svc at all.
-func (und undeleter) newestMtime(logger logrus.FieldLogger, blk string, svc arvados.KeepService) (time.Time, error) {
- found, err := svc.Index(und.client, blk)
+func (rcvr recoverer) newestMtime(logger logrus.FieldLogger, blk string, svc arvados.KeepService) (time.Time, error) {
+ found, err := svc.Index(rcvr.client, blk)
if err != nil {
logger.WithError(err).Warn("error getting index")
return time.Time{}, err
// decide to trash it, all before our recovered collection gets
// saved. But if the block's timestamp is more recent than blobsigttl,
// keepstore will refuse to trash it even if told to by keep-balance.
-func (und undeleter) ensureSafe(ctx context.Context, logger logrus.FieldLogger, blk string, svc arvados.KeepService, blobsigttl time.Duration, blobsigexp time.Time) error {
- if latest, err := und.newestMtime(logger, blk, svc); err != nil {
+func (rcvr recoverer) ensureSafe(ctx context.Context, logger logrus.FieldLogger, blk string, svc arvados.KeepService, blobsigttl time.Duration, blobsigexp time.Time) error {
+ if latest, err := rcvr.newestMtime(logger, blk, svc); err != nil {
return err
} else if latest.Add(blobsigttl).After(blobsigexp) {
return nil
}
- if err := svc.Touch(ctx, und.client, blk); err != nil {
+ if err := svc.Touch(ctx, rcvr.client, blk); err != nil {
return fmt.Errorf("error updating timestamp: %s", err)
}
logger.Debug("updated timestamp")
- if latest, err := und.newestMtime(logger, blk, svc); err == errNotFound {
+ if latest, err := rcvr.newestMtime(logger, blk, svc); err == errNotFound {
return fmt.Errorf("(BUG?) touch succeeded, but then block did not appear in index")
} else if err != nil {
return err
// Untrash and update GC timestamps (as needed) on blocks referenced
// by the given manifest, save a new collection and return the new
// collection's UUID.
-func (und undeleter) RecoverManifest(mtxt string) (string, error) {
+func (rcvr recoverer) RecoverManifest(mtxt string) (string, error) {
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
go close(todo)
var services []arvados.KeepService
- err = und.client.EachKeepService(func(svc arvados.KeepService) error {
+ err = rcvr.client.EachKeepService(func(svc arvados.KeepService) error {
if svc.ServiceType == "proxy" {
- und.logger.WithField("service", svc).Debug("ignore proxy service")
+ rcvr.logger.WithField("service", svc).Debug("ignore proxy service")
} else {
services = append(services, svc)
}
if err != nil {
return "", fmt.Errorf("error getting list of keep services: %s", err)
}
- und.logger.WithField("services", services).Debug("got list of services")
+ rcvr.logger.WithField("services", services).Debug("got list of services")
// blobsigexp is our deadline for saving the rescued
// collection. This must be less than BlobSigningTTL
// would have lived long enough anyway if left alone.
// BlobSigningTTL/2 (typically around 1 week) is much longer
// than than we need to recover even a very large collection.
- blobsigttl := und.cluster.Collections.BlobSigningTTL.Duration()
+ blobsigttl := rcvr.cluster.Collections.BlobSigningTTL.Duration()
blobsigexp := time.Now().Add(blobsigttl / 2)
- und.logger.WithField("blobsigexp", blobsigexp).Debug("chose save deadline")
+ rcvr.logger.WithField("blobsigexp", blobsigexp).Debug("chose save deadline")
// We'll start a number of threads, each working on
// checking/recovering one block at a time. The threads
nextblk:
for idx := range todo {
blk := strings.SplitN(string(blks[idx]), "+", 2)[0]
- logger := und.logger.WithField("block", blk)
+ logger := rcvr.logger.WithField("block", blk)
for _, untrashing := range []bool{false, true} {
for _, svc := range services {
logger := logger.WithField("service", fmt.Sprintf("%s:%d", svc.ServiceHost, svc.ServicePort))
if untrashing {
- if err := svc.Untrash(ctx, und.client, blk); err != nil {
+ if err := svc.Untrash(ctx, rcvr.client, blk); err != nil {
logger.WithError(err).Debug("untrash failed")
continue
}
logger.Info("untrashed")
}
- err := und.ensureSafe(ctx, logger, blk, svc, blobsigttl, blobsigexp)
+ err := rcvr.ensureSafe(ctx, logger, blk, svc, blobsigttl, blobsigexp)
if err == errNotFound {
logger.Debug(err)
} else if err != nil {
}
if havenot > 0 {
if have > 0 {
- und.logger.Warn("partial recovery is not implemented")
+ rcvr.logger.Warn("partial recovery is not implemented")
}
return "", fmt.Errorf("unable to recover %d of %d blocks", havenot, have+havenot)
}
- if und.cluster.Collections.BlobSigning {
- key := []byte(und.cluster.Collections.BlobSigningKey)
- coll.ManifestText = arvados.SignManifest(coll.ManifestText, und.client.AuthToken, blobsigexp, blobsigttl, key)
+ if rcvr.cluster.Collections.BlobSigning {
+ key := []byte(rcvr.cluster.Collections.BlobSigningKey)
+ coll.ManifestText = arvados.SignManifest(coll.ManifestText, rcvr.client.AuthToken, blobsigexp, blobsigttl, key)
}
- und.logger.WithField("manifest", coll.ManifestText).Debug("updated blob signatures in manifest")
- err = und.client.RequestAndDecodeContext(ctx, &coll, "POST", "arvados/v1/collections", nil, map[string]interface{}{
+ rcvr.logger.WithField("manifest", coll.ManifestText).Debug("updated blob signatures in manifest")
+ err = rcvr.client.RequestAndDecodeContext(ctx, &coll, "POST", "arvados/v1/collections", nil, map[string]interface{}{
"collection": map[string]interface{}{
"manifest_text": coll.ManifestText,
},
if err != nil {
return "", fmt.Errorf("error saving new collection: %s", err)
}
- und.logger.WithField("UUID", coll.UUID).Debug("created new collection")
+ rcvr.logger.WithField("UUID", coll.UUID).Debug("created new collection")
return coll.UUID, nil
}
//
// SPDX-License-Identifier: AGPL-3.0
-package undelete
+package recovercollection
import (
"bytes"
mfile := tmp + "/manifest"
ioutil.WriteFile(mfile, []byte(". aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa+410 0:410:Gone\n"), 0777)
var stdout, stderr bytes.Buffer
- exitcode := Command.RunCommand("undelete.test", []string{"-log-level=debug", mfile}, &bytes.Buffer{}, &stdout, &stderr)
+ exitcode := Command.RunCommand("recovercollection.test", []string{"-log-level=debug", mfile}, &bytes.Buffer{}, &stdout, &stderr)
c.Check(exitcode, check.Equals, 1)
c.Check(stdout.String(), check.Equals, "")
c.Log(stderr.String())
}
var stdout, stderr bytes.Buffer
- exitcode := Command.RunCommand("undelete.test", []string{"-log-level=debug", mfile}, &bytes.Buffer{}, &stdout, &stderr)
+ exitcode := Command.RunCommand("recovercollection.test", []string{"-log-level=debug", mfile}, &bytes.Buffer{}, &stdout, &stderr)
c.Check(exitcode, check.Equals, 0)
c.Check(stdout.String(), check.Matches, `zzzzz-4zz18-.{15}\n`)
c.Log(stderr.String())
}
c.Check(found, check.Equals, true)
}
+
+func (*Suite) TestUnusableManifestSourceArg(c *check.C) {
+ for _, trial := range []struct {
+ srcArg string
+ errRegexp string
+ }{
+ {"zzzzz-4zz18-aaaaaaaaaaaaaaa", `(?ms).*msg="log entry not found".*`},
+ {"zzzzz-57u5n-aaaaaaaaaaaaaaa", `(?ms).*msg="log entry not found.*`},
+ {"zzzzz-57u5n-containerlog006", `(?ms).*msg="log entry properties\.old_attributes\.manifest_text missing or empty".*`},
+ {"zzzzz-j7d0g-aaaaaaaaaaaaaaa", `(?ms).*msg="looks like a UUID but not a log or collection UUID.*`},
+ } {
+ var stdout, stderr bytes.Buffer
+ exitcode := Command.RunCommand("recovercollection.test", []string{"-log-level=debug", trial.srcArg}, &bytes.Buffer{}, &stdout, &stderr)
+ c.Check(exitcode, check.Equals, 1)
+ c.Check(stdout.String(), check.Equals, "")
+ c.Log(stderr.String())
+ c.Check(stderr.String(), check.Matches, trial.errRegexp)
+ }
+}
import cwltool.process
import cwltool.argparser
from cwltool.process import shortname, UnsupportedRequirement, use_custom_schema
-from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, get_listing
+from cwltool.utils import adjustFileObjs, adjustDirObjs, get_listing
import arvados
import arvados.config
cwltool.command_line_tool.ACCEPTLIST_RE = cwltool.command_line_tool.ACCEPTLIST_EN_RELAXED_RE
res10 = pkg_resources.resource_stream(__name__, 'arv-cwl-schema-v1.0.yml')
res11 = pkg_resources.resource_stream(__name__, 'arv-cwl-schema-v1.1.yml')
- customschema10 = res10.read()
- customschema11 = res11.read()
+ customschema10 = res10.read().decode('utf-8')
+ customschema11 = res11.read().decode('utf-8')
use_custom_schema("v1.0", "http://arvados.org/cwl", customschema10)
use_custom_schema("v1.1.0-dev1", "http://arvados.org/cwl", customschema11)
use_custom_schema("v1.1", "http://arvados.org/cwl", customschema11)
from ._version import __version__
from cwltool.process import shortname, UnsupportedRequirement, use_custom_schema
-from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, get_listing, visit_class
+from cwltool.utils import adjustFileObjs, adjustDirObjs, get_listing, visit_class
from cwltool.command_line_tool import compute_checksums
from cwltool.load_tool import load_tool
# file to determine what version of cwltool and schema-salad to
# build.
install_requires=[
- 'cwltool==3.0.20200324120055',
- 'schema-salad==5.0.20200302192450',
+ 'cwltool==3.0.20200530110633',
+ 'schema-salad==6.0.20200601095207',
'arvados-python-client{}'.format(pysdk_dep),
'setuptools',
'ciso8601 >= 2.0.0'
AlternateEmailAddresses bool
}
OpenIDConnect struct {
- Enable bool
- Issuer string
- ClientID string
- ClientSecret string
+ Enable bool
+ Issuer string
+ ClientID string
+ ClientSecret string
+ EmailClaim string
+ EmailVerifiedClaim string
+ UsernameClaim string
}
PAM struct {
Enable bool
'ruamel.yaml >=0.15.54, <=0.16.5',
'setuptools',
'ws4py >=0.4.2',
+ 'rsa < 4.1'
],
extras_require={
':os.name=="posix" and python_version<"3"': ['subprocess32 >= 3.5.1'],
raise
end
- require 'refresh_permission_view'
+ require 'update_permissions'
- refresh_permission_view
+ refresh_permissions
+ refresh_trashed
# Done.
send_json success: true
sql_conds = nil
user_uuids = users_list.map { |u| u.uuid }
+ # For details on how the trashed_groups table is constructed, see
+ # see db/migrate/20200501150153_permission_table.rb
+
exclude_trashed_records = ""
if !include_trash and (sql_table == "groups" or sql_table == "collections") then
- # Only include records that are not explicitly trashed
- exclude_trashed_records = "AND #{sql_table}.is_trashed = false"
+ # Only include records that are not trashed
+ exclude_trashed_records = "AND (#{sql_table}.trash_at is NULL or #{sql_table}.trash_at > statement_timestamp())"
end
if users_list.select { |u| u.is_admin }.any?
if !include_trash
if sql_table != "api_client_authorizations"
# Only include records where the owner is not trashed
- sql_conds = "#{sql_table}.owner_uuid NOT IN (SELECT target_uuid FROM #{PERMISSION_VIEW} "+
- "WHERE trashed = 1) #{exclude_trashed_records}"
+ sql_conds = "#{sql_table}.owner_uuid NOT IN (SELECT group_uuid FROM #{TRASHED_GROUPS} "+
+ "where trash_at <= statement_timestamp()) #{exclude_trashed_records}"
end
end
else
trashed_check = ""
if !include_trash then
- trashed_check = "AND trashed = 0"
+ trashed_check = "AND target_uuid NOT IN (SELECT group_uuid FROM #{TRASHED_GROUPS} where trash_at <= statement_timestamp())"
end
+ # The core of the permission check is a join against the
+ # materialized_permissions table to determine if the user has at
+ # least read permission to either the object itself or its
+ # direct owner (if traverse_owned is true). See
+ # db/migrate/20200501150153_permission_table.rb for details on
+ # how the permissions are computed.
+
+ # A user can have can_manage access to another user, this grants
+ # full access to all that user's stuff. To implement that we
+ # need to include those other users in the permission query.
+ user_uuids_subquery = USER_UUIDS_SUBQUERY_TEMPLATE % {user: ":user_uuids", perm_level: 1}
+
# Note: it is possible to combine the direct_check and
# owner_check into a single EXISTS() clause, however it turns
# out query optimizer doesn't like it and forces a sequential
# Match a direct read permission link from the user to the record uuid
direct_check = "#{sql_table}.uuid IN (SELECT target_uuid FROM #{PERMISSION_VIEW} "+
- "WHERE user_uuid IN (:user_uuids) AND perm_level >= 1 #{trashed_check})"
+ "WHERE user_uuid IN (#{user_uuids_subquery}) AND perm_level >= 1 #{trashed_check})"
- # Match a read permission link from the user to the record's owner_uuid
+ # Match a read permission for the user to the record's
+ # owner_uuid. This is so we can have a permissions table that
+ # mostly consists of users and groups (projects are a type of
+ # group) and not have to compute and list user permission to
+ # every single object in the system.
+ #
+ # Don't do this for API keys (special behavior) or groups
+ # (already covered by direct_check).
+ #
+ # The traverse_owned flag indicates whether the permission to
+ # read an object also implies transitive permission to read
+ # things the object owns. The situation where this is important
+ # are determining if we can read an object owned by another
+ # user. This makes it possible to have permission to read the
+ # user record without granting permission to read things the
+ # other user owns.
owner_check = ""
if sql_table != "api_client_authorizations" and sql_table != "groups" then
owner_check = "OR #{sql_table}.owner_uuid IN (SELECT target_uuid FROM #{PERMISSION_VIEW} "+
- "WHERE user_uuid IN (:user_uuids) AND perm_level >= 1 #{trashed_check} AND target_owner_uuid IS NOT NULL) "
+ "WHERE user_uuid IN (#{user_uuids_subquery}) AND perm_level >= 1 #{trashed_check} AND traverse_owned) "
end
links_cond = ""
# users some permission _or_ gives anyone else permission to
# view one of the authorized users.
links_cond = "OR (#{sql_table}.link_class IN (:permission_link_classes) AND "+
- "(#{sql_table}.head_uuid IN (:user_uuids) OR #{sql_table}.tail_uuid IN (:user_uuids)))"
+ "(#{sql_table}.head_uuid IN (#{user_uuids_subquery}) OR #{sql_table}.tail_uuid IN (#{user_uuids_subquery})))"
end
sql_conds = "(#{direct_check} #{owner_check} #{links_cond}) #{exclude_trashed_records}"
#
# SPDX-License-Identifier: AGPL-3.0
+require 'update_permissions'
+
class DatabaseSeeds
extend CurrentApiClient
def self.install
anonymous_group_read_permission
anonymous_user
empty_collection
+ refresh_permissions
+ refresh_trashed
end
end
attribute :properties, :jsonbHash, default: {}
validate :ensure_filesystem_compatible_name
- after_create :invalidate_permissions_cache
- after_update :maybe_invalidate_permissions_cache
before_create :assign_name
+ after_create :after_ownership_change
+ after_create :update_trash
+
+ before_update :before_ownership_change
+ after_update :after_ownership_change
+
+ after_update :update_trash
+ before_destroy :clear_permissions_and_trash
api_accessible :user, extend: :common do |t|
t.add :name
super if group_class == 'project'
end
- def maybe_invalidate_permissions_cache
- if uuid_changed? or owner_uuid_changed? or is_trashed_changed?
- # This can change users' permissions on other groups as well as
- # this one.
- invalidate_permissions_cache
+ def update_trash
+ if trash_at_changed? or owner_uuid_changed?
+ # The group was added or removed from the trash.
+ #
+ # Strategy:
+ # Compute project subtree, propagating trash_at to subprojects
+ # Remove groups that don't belong from trash
+ # Add/update groups that do belong in the trash
+
+ temptable = "group_subtree_#{rand(2**64).to_s(10)}"
+ ActiveRecord::Base.connection.exec_query %{
+create temporary table #{temptable} on commit drop
+as select * from project_subtree_with_trash_at($1, LEAST($2, $3)::timestamp)
+},
+ 'Group.update_trash.select',
+ [[nil, self.uuid],
+ [nil, TrashedGroup.find_by_group_uuid(self.owner_uuid).andand.trash_at],
+ [nil, self.trash_at]]
+
+ ActiveRecord::Base.connection.exec_delete %{
+delete from trashed_groups where group_uuid in (select target_uuid from #{temptable} where trash_at is NULL);
+},
+ "Group.update_trash.delete"
+
+ ActiveRecord::Base.connection.exec_query %{
+insert into trashed_groups (group_uuid, trash_at)
+ select target_uuid as group_uuid, trash_at from #{temptable} where trash_at is not NULL
+on conflict (group_uuid) do update set trash_at=EXCLUDED.trash_at;
+},
+ "Group.update_trash.insert"
+ end
+ end
+
+ def before_ownership_change
+ if owner_uuid_changed? and !self.owner_uuid_was.nil?
+ MaterializedPermission.where(user_uuid: owner_uuid_was, target_uuid: uuid).delete_all
+ update_permissions self.owner_uuid_was, self.uuid, REVOKE_PERM
+ end
+ end
+
+ def after_ownership_change
+ if owner_uuid_changed?
+ update_permissions self.owner_uuid, self.uuid, CAN_MANAGE_PERM
end
end
- def invalidate_permissions_cache
- # Ensure a new group can be accessed by the appropriate users
- # immediately after being created.
- User.invalidate_permissions_cache self.async_permissions_update
+ def clear_permissions_and_trash
+ MaterializedPermission.where(target_uuid: uuid).delete_all
+ ActiveRecord::Base.connection.exec_delete %{
+delete from trashed_groups where group_uuid=$1
+}, "Group.clear_permissions_and_trash", [[nil, self.uuid]]
+
end
def assign_name
# already know how to properly treat them.
attribute :properties, :jsonbHash, default: {}
+ validate :name_links_are_obsolete
before_create :permission_to_attach_to_objects
before_update :permission_to_attach_to_objects
- after_update :maybe_invalidate_permissions_cache
- after_create :maybe_invalidate_permissions_cache
- after_destroy :maybe_invalidate_permissions_cache
- validate :name_links_are_obsolete
+ after_update :call_update_permissions
+ after_create :call_update_permissions
+ before_destroy :clear_permissions
+ after_destroy :check_permissions
api_accessible :user, extend: :common do |t|
t.add :tail_uuid
false
end
- def maybe_invalidate_permissions_cache
+ PERM_LEVEL = {
+ 'can_read' => 1,
+ 'can_login' => 1,
+ 'can_write' => 2,
+ 'can_manage' => 3,
+ }
+
+ def call_update_permissions
+ if self.link_class == 'permission'
+ update_permissions tail_uuid, head_uuid, PERM_LEVEL[name], self.uuid
+ end
+ end
+
+ def clear_permissions
+ if self.link_class == 'permission'
+ update_permissions tail_uuid, head_uuid, REVOKE_PERM, self.uuid
+ end
+ end
+
+ def check_permissions
if self.link_class == 'permission'
- # Clearing the entire permissions cache can generate many
- # unnecessary queries if many active users are not affected by
- # this change. In such cases it would be better to search cached
- # permissions for head_uuid and tail_uuid, and invalidate the
- # cache for only those users. (This would require a browseable
- # cache.)
- User.invalidate_permissions_cache
+ check_permissions_against_full_refresh
end
end
--- /dev/null
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+class MaterializedPermission < ApplicationRecord
+end
--- /dev/null
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+class TrashedGroup < ApplicationRecord
+end
# SPDX-License-Identifier: AGPL-3.0
require 'can_be_an_owner'
-require 'refresh_permission_view'
class User < ArvadosModel
include HasUuid
user.username.nil? and user.username_changed?
}
before_update :setup_on_activate
+
before_create :check_auto_admin
before_create :set_initial_username, :if => Proc.new { |user|
user.username.nil? and user.email
}
+ after_create :after_ownership_change
after_create :setup_on_activate
after_create :add_system_group_permission_link
- after_create :invalidate_permissions_cache
after_create :auto_setup_new_user, :if => Proc.new { |user|
Rails.configuration.Users.AutoSetupNewUsers and
(user.uuid != system_user_uuid) and
(user.uuid != anonymous_user_uuid)
}
after_create :send_admin_notifications
+
+ before_update :before_ownership_change
+ after_update :after_ownership_change
after_update :send_profile_created_notification
after_update :sync_repository_names, :if => Proc.new { |user|
(user.uuid != system_user_uuid) and
user.username_changed? and
(not user.username_was.nil?)
}
+ before_destroy :clear_permissions
+ after_destroy :remove_self_from_permissions
has_many :authorized_keys, :foreign_key => :authorized_user_uuid, :primary_key => :uuid
has_many :repositories, foreign_key: :owner_uuid, primary_key: :uuid
{read: true, write: true},
{read: true, write: true, manage: true}]
+ VAL_FOR_PERM =
+ {:read => 1,
+ :write => 2,
+ :manage => 3}
+
+
def full_name
"#{first_name} #{last_name}".strip
end
end
def groups_i_can(verb)
- my_groups = self.group_permissions.select { |uuid, mask| mask[verb] }.keys
+ my_groups = self.group_permissions(VAL_FOR_PERM[verb]).keys
if verb == :read
my_groups << anonymous_group_uuid
end
end
end
next if target_uuid == self.uuid
- next if (group_permissions[target_uuid] and
- group_permissions[target_uuid][action])
- if target.respond_to? :owner_uuid
- next if target.owner_uuid == self.uuid
- next if (group_permissions[target.owner_uuid] and
- group_permissions[target.owner_uuid][action])
- end
- sufficient_perms = case action
- when :manage
- ['can_manage']
- when :write
- ['can_manage', 'can_write']
- when :read
- ['can_manage', 'can_write', 'can_read']
- else
- # (Skip this kind of permission opportunity
- # if action is an unknown permission type)
- end
- if sufficient_perms
- # Check permission links with head_uuid pointing directly at
- # the target object. If target is a Group, this is redundant
- # and will fail except [a] if permission caching is broken or
- # [b] during a race condition, where a permission link has
- # *just* been added.
- if Link.where(link_class: 'permission',
- name: sufficient_perms,
- tail_uuid: groups_i_can(action) + [self.uuid],
- head_uuid: target_uuid).any?
- next
- end
+
+ target_owner_uuid = target.owner_uuid if target.respond_to? :owner_uuid
+
+ user_uuids_subquery = USER_UUIDS_SUBQUERY_TEMPLATE % {user: "$1", perm_level: "$3"}
+
+ unless ActiveRecord::Base.connection.
+ exec_query(%{
+SELECT 1 FROM #{PERMISSION_VIEW}
+ WHERE user_uuid in (#{user_uuids_subquery}) and
+ ((target_uuid = $2 and perm_level >= $3)
+ or (target_uuid = $4 and perm_level >= $3 and traverse_owned))
+},
+ # "name" arg is a query label that appears in logs:
+ "user_can_query",
+ [[nil, self.uuid],
+ [nil, target_uuid],
+ [nil, VAL_FOR_PERM[action]],
+ [nil, target_owner_uuid]]
+ ).any?
+ return false
end
- return false
end
true
end
- def self.invalidate_permissions_cache(async=false)
- refresh_permission_view(async)
+ def before_ownership_change
+ if owner_uuid_changed? and !self.owner_uuid_was.nil?
+ MaterializedPermission.where(user_uuid: owner_uuid_was, target_uuid: uuid).delete_all
+ update_permissions self.owner_uuid_was, self.uuid, REVOKE_PERM
+ end
+ end
+
+ def after_ownership_change
+ if owner_uuid_changed?
+ update_permissions self.owner_uuid, self.uuid, CAN_MANAGE_PERM
+ end
+ end
+
+ def clear_permissions
+ MaterializedPermission.where("user_uuid = ? and target_uuid != ?", uuid, uuid).delete_all
end
- def invalidate_permissions_cache
- User.invalidate_permissions_cache
+ def remove_self_from_permissions
+ MaterializedPermission.where("target_uuid = ?", uuid).delete_all
+ check_permissions_against_full_refresh
end
# Return a hash of {user_uuid: group_perms}
+ #
+ # note: this does not account for permissions that a user gains by
+ # having can_manage on another user.
def self.all_group_permissions
all_perms = {}
ActiveRecord::Base.connection.
- exec_query("SELECT user_uuid, target_owner_uuid, perm_level, trashed
+ exec_query(%{
+SELECT user_uuid, target_uuid, perm_level
FROM #{PERMISSION_VIEW}
- WHERE target_owner_uuid IS NOT NULL",
+ WHERE traverse_owned
+},
# "name" arg is a query label that appears in logs:
- "all_group_permissions",
- ).rows.each do |user_uuid, group_uuid, max_p_val, trashed|
+ "all_group_permissions").
+ rows.each do |user_uuid, group_uuid, max_p_val|
all_perms[user_uuid] ||= {}
all_perms[user_uuid][group_uuid] = PERMS_FOR_VAL[max_p_val.to_i]
end
# Return a hash of {group_uuid: perm_hash} where perm_hash[:read]
# and perm_hash[:write] are true if this user can read and write
# objects owned by group_uuid.
- def group_permissions
- group_perms = {self.uuid => {:read => true, :write => true, :manage => true}}
+ def group_permissions(level=1)
+ group_perms = {}
+
+ user_uuids_subquery = USER_UUIDS_SUBQUERY_TEMPLATE % {user: "$1", perm_level: "$2"}
+
ActiveRecord::Base.connection.
- exec_query("SELECT target_owner_uuid, perm_level, trashed
- FROM #{PERMISSION_VIEW}
- WHERE user_uuid = $1
- AND target_owner_uuid IS NOT NULL",
+ exec_query(%{
+SELECT target_uuid, perm_level
+ FROM #{PERMISSION_VIEW}
+ WHERE user_uuid in (#{user_uuids_subquery}) and perm_level >= $2
+},
# "name" arg is a query label that appears in logs:
- "group_permissions for #{uuid}",
+ "User.group_permissions",
# "binds" arg is an array of [col_id, value] for '$1' vars:
- [[nil, uuid]],
- ).rows.each do |group_uuid, max_p_val, trashed|
+ [[nil, uuid],
+ [nil, level]]).
+ rows.each do |group_uuid, max_p_val|
group_perms[group_uuid] = PERMS_FOR_VAL[max_p_val.to_i]
end
group_perms
self.uuid = new_uuid
save!(validate: false)
change_all_uuid_refs(old_uuid: old_uuid, new_uuid: new_uuid)
+ ActiveRecord::Base.connection.exec_update %{
+update #{PERMISSION_VIEW} set user_uuid=$1 where user_uuid = $2
+},
+ 'User.update_uuid.update_permissions_user_uuid',
+ [[nil, new_uuid],
+ [nil, old_uuid]]
+ ActiveRecord::Base.connection.exec_update %{
+update #{PERMISSION_VIEW} set target_uuid=$1 where target_uuid = $2
+},
+ 'User.update_uuid.update_permissions_target_uuid',
+ [[nil, new_uuid],
+ [nil, old_uuid]]
end
end
raise "user does not exist" if !new_user
raise "cannot merge to an already merged user" if new_user.redirect_to_user_uuid
+ self.clear_permissions
+ new_user.clear_permissions
+
# If 'self' is a remote user, don't transfer authorizations
# (i.e. ability to access the account) to the new user, because
# that gives the remote site the ability to access the 'new'
if redirect_to_new_user
update_attributes!(redirect_to_user_uuid: new_user.uuid, username: nil)
end
- invalidate_permissions_cache
+ skip_check_permissions_against_full_refresh do
+ update_permissions self.uuid, self.uuid, CAN_MANAGE_PERM
+ update_permissions new_user.uuid, new_user.uuid, CAN_MANAGE_PERM
+ update_permissions new_user.owner_uuid, new_user.uuid, CAN_MANAGE_PERM
+ end
+ update_permissions self.owner_uuid, self.uuid, CAN_MANAGE_PERM
end
end
--- /dev/null
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+require '20200501150153_permission_table_constants'
+
+class PermissionTable < ActiveRecord::Migration[5.0]
+ def up
+ # This is a major migration. We are replacing the
+ # materialized_permission_view, which is fully recomputed any time
+ # a permission changes (and becomes very expensive as the number
+ # of users/groups becomes large), with a new strategy that only
+ # recomputes permissions for the subset of objects that are
+ # potentially affected by the addition or removal of a permission
+ # relationship (i.e. ownership or a permission link).
+ #
+ # This also disentangles the concept of "trashed groups" from the
+ # permissions system. Updating trashed items follows a similar
+ # (but less complicated) strategy to updating permissions, so it
+ # may be helpful to look at that first.
+
+ ActiveRecord::Base.connection.execute "DROP MATERIALIZED VIEW IF EXISTS materialized_permission_view;"
+ drop_table :permission_refresh_lock
+
+ # This table stores the set of trashed groups and their trash_at
+ # time. Used to exclude trashed projects and their contents when
+ # getting object listings.
+ create_table :trashed_groups, :id => false do |t|
+ t.string :group_uuid
+ t.datetime :trash_at
+ end
+ add_index :trashed_groups, :group_uuid, :unique => true
+
+ ActiveRecord::Base.connection.execute %{
+create or replace function project_subtree_with_trash_at (starting_uuid varchar(27), starting_trash_at timestamp)
+returns table (target_uuid varchar(27), trash_at timestamp)
+STABLE
+language SQL
+as $$
+/* Starting from a project, recursively traverse all the projects
+ underneath it and return a set of project uuids and trash_at times
+ (may be null). The initial trash_at can be a timestamp or null.
+ The trash_at time propagates downward to groups it owns, i.e. when a
+ group is trashed, everything underneath it in the ownership
+ hierarchy is also considered trashed. However, this is fact is
+ recorded in the trashed_groups table, not by updating trash_at field
+ in the groups table.
+*/
+WITH RECURSIVE
+ project_subtree(uuid, trash_at) as (
+ values (starting_uuid, starting_trash_at)
+ union
+ select groups.uuid, LEAST(project_subtree.trash_at, groups.trash_at)
+ from groups join project_subtree on (groups.owner_uuid = project_subtree.uuid)
+ )
+ select uuid, trash_at from project_subtree;
+$$;
+}
+
+ # Now populate the table. For a non-test databse this is the only
+ # time this ever happens, after this the trash table is updated
+ # incrementally. See app/models/group.rb#update_trash
+ refresh_trashed
+
+ # The table to store the flattened permissions. This is almost
+ # exactly the same as the old materalized_permission_view except
+ # that the target_owner_uuid colunm in the view is now just a
+ # boolean traverse_owned (the column was only ever tested for null
+ # or non-null).
+ #
+ # For details on how this table is used to apply permissions to
+ # queries, see app/models/arvados_model.rb#readable_by
+ #
+ create_table :materialized_permissions, :id => false do |t|
+ t.string :user_uuid
+ t.string :target_uuid
+ t.integer :perm_level
+ t.boolean :traverse_owned
+ end
+ add_index :materialized_permissions, [:user_uuid, :target_uuid], unique: true, name: 'permission_user_target'
+ add_index :materialized_permissions, [:target_uuid], unique: false, name: 'permission_target'
+
+ ActiveRecord::Base.connection.execute %{
+create or replace function should_traverse_owned (starting_uuid varchar(27),
+ starting_perm integer)
+ returns bool
+IMMUTABLE
+language SQL
+as $$
+/* Helper function. Determines if permission on an object implies
+ transitive permission to things the object owns. This is always
+ true for groups, but only true for users when the permission level
+ is can_manage.
+*/
+select starting_uuid like '_____-j7d0g-_______________' or
+ (starting_uuid like '_____-tpzed-_______________' and starting_perm >= 3);
+$$;
+}
+
+ # Merge all permission relationships into a single view. This
+ # consists of: groups owned by users and projects, users owned
+ # by other users, users have permission on themselves,
+ # and explicit permission links.
+ #
+ # A SQL view gets inlined into the query where it is used as a
+ # subquery. This enables the query planner to inject constraints,
+ # so it only has to look up edges it plans to traverse and avoid a brute
+ # force query of all edges.
+ ActiveRecord::Base.connection.execute %{
+create view permission_graph_edges as
+ select groups.owner_uuid as tail_uuid, groups.uuid as head_uuid,
+ (3) as val, groups.uuid as edge_id from groups
+union all
+ select users.owner_uuid as tail_uuid, users.uuid as head_uuid,
+ (3) as val, users.uuid as edge_id from users
+union all
+ select users.uuid as tail_uuid, users.uuid as head_uuid,
+ (3) as val, '' as edge_id from users
+union all
+ select links.tail_uuid,
+ links.head_uuid,
+ CASE
+ WHEN links.name = 'can_read' THEN 1
+ WHEN links.name = 'can_login' THEN 1
+ WHEN links.name = 'can_write' THEN 2
+ WHEN links.name = 'can_manage' THEN 3
+ ELSE 0
+ END as val,
+ links.uuid as edge_id
+ from links
+ where links.link_class='permission'
+}
+
+ # This is used to ensure that the permission edge passed into
+ # compute_permission_subgraph takes replaces the existing edge in
+ # the "edges" view that is about to be removed.
+ edge_perm = %{
+case (edges.edge_id = perm_edge_id)
+ when true then starting_perm
+ else edges.val
+ end
+}
+
+ # The primary function to compute permissions for a subgraph.
+ # Comments on how it works are inline.
+ #
+ # Due to performance issues due to the query optimizer not
+ # working across function and "with" expression boundaries, I
+ # had to fall back on using string templates for repeated code
+ # in order to inline it.
+
+ ActiveRecord::Base.connection.execute %{
+create or replace function compute_permission_subgraph (perm_origin_uuid varchar(27),
+ starting_uuid varchar(27),
+ starting_perm integer,
+ perm_edge_id varchar(27))
+returns table (user_uuid varchar(27), target_uuid varchar(27), val integer, traverse_owned bool)
+STABLE
+language SQL
+as $$
+
+/* The purpose of this function is to compute the permissions for a
+ subgraph of the database, starting from a given edge. The newly
+ computed permissions are used to add and remove rows from the main
+ permissions table.
+
+ perm_origin_uuid: The object that 'gets' the permission.
+
+ starting_uuid: The starting object the permission applies to.
+
+ starting_perm: The permission that perm_origin_uuid 'has' on
+ starting_uuid One of 1, 2, 3 for can_read,
+ can_write, can_manage respectively, or 0 to revoke
+ permissions.
+
+ perm_edge_id: Identifies the permission edge that is being updated.
+ Changes of ownership, this is starting_uuid.
+ For links, this is the uuid of the link object.
+ This is used to override the edge value in the database
+ with starting_perm. This is necessary when revoking
+ permissions because the update happens before edge is
+ actually removed.
+*/
+with
+ /* Starting from starting_uuid, determine the set of objects that
+ could be affected by this permission change.
+
+ Note: We don't traverse users unless it is an "identity"
+ permission (permission origin is self).
+ */
+ perm_from_start(perm_origin_uuid, target_uuid, val, traverse_owned) as (
+ #{PERM_QUERY_TEMPLATE % {:base_case => %{
+ values (perm_origin_uuid, starting_uuid, starting_perm,
+ should_traverse_owned(starting_uuid, starting_perm),
+ (perm_origin_uuid = starting_uuid or starting_uuid not like '_____-tpzed-_______________'))
+},
+:edge_perm => edge_perm
+} }),
+
+ /* Find other inbound edges that grant permissions to 'targets' in
+ perm_from_start, and compute permissions that originate from
+ those.
+
+ This is necessary for two reasons:
+
+ 1) Other users may have access to a subset of the objects
+ through other permission links than the one we started from.
+ If we don't recompute them, their permission will get dropped.
+
+ 2) There may be more than one path through which a user gets
+ permission to an object. For example, a user owns a project
+ and also shares it can_read with a group the user belongs
+ to. adding the can_read link must not overwrite the existing
+ can_manage permission granted by ownership.
+ */
+ additional_perms(perm_origin_uuid, target_uuid, val, traverse_owned) as (
+ #{PERM_QUERY_TEMPLATE % {:base_case => %{
+ select edges.tail_uuid as origin_uuid, edges.head_uuid as target_uuid, edges.val,
+ should_traverse_owned(edges.head_uuid, edges.val),
+ edges.head_uuid like '_____-j7d0g-_______________'
+ from permission_graph_edges as edges
+ where edges.edge_id != perm_edge_id and
+ edges.tail_uuid not in (select target_uuid from perm_from_start where target_uuid like '_____-j7d0g-_______________') and
+ edges.head_uuid in (select target_uuid from perm_from_start)
+},
+:edge_perm => edge_perm
+} }),
+
+ /* Combine the permissions computed in the first two phases. */
+ all_perms(perm_origin_uuid, target_uuid, val, traverse_owned) as (
+ select * from perm_from_start
+ union all
+ select * from additional_perms
+ )
+
+ /* The actual query that produces rows to be added or removed
+ from the materialized_permissions table. This is the clever
+ bit.
+
+ Key insights:
+
+ * For every group, the materialized_permissions lists all users
+ that can access to that group.
+
+ * The all_perms subquery has computed permissions on on a set of
+ objects for all inbound "origins", which are users or groups.
+
+ * Permissions through groups are transitive.
+
+ We can infer:
+
+ 1) The materialized_permissions table declares that user X has permission N on group Y
+ 2) The all_perms result has determined group Y has permission M on object Z
+ 3) Therefore, user X has permission min(N, M) on object Z
+
+ This allows us to efficiently determine the set of users that
+ have permissions on the subset of objects, without having to
+ follow the chain of permission back up to find those users.
+
+ In addition, because users always have permission on themselves, this
+ query also makes sure those permission rows are always
+ returned.
+ */
+ select v.user_uuid, v.target_uuid, max(v.perm_level), bool_or(v.traverse_owned) from
+ (select m.user_uuid,
+ u.target_uuid,
+ least(u.val, m.perm_level) as perm_level,
+ u.traverse_owned
+ from all_perms as u, materialized_permissions as m
+ where u.perm_origin_uuid = m.target_uuid AND m.traverse_owned
+ AND (m.user_uuid = m.target_uuid or m.target_uuid not like '_____-tpzed-_______________')
+ union all
+ select target_uuid as user_uuid, target_uuid, 3, true
+ from all_perms
+ where all_perms.target_uuid like '_____-tpzed-_______________') as v
+ group by v.user_uuid, v.target_uuid
+$$;
+ }
+
+ #
+ # Populate materialized_permissions by traversing permissions
+ # starting at each user.
+ #
+ refresh_permissions
+ end
+
+ def down
+ drop_table :materialized_permissions
+ drop_table :trashed_groups
+
+ ActiveRecord::Base.connection.execute "DROP function project_subtree_with_trash_at (varchar, timestamp);"
+ ActiveRecord::Base.connection.execute "DROP function compute_permission_subgraph (varchar, varchar, integer, varchar);"
+ ActiveRecord::Base.connection.execute "DROP function should_traverse_owned(varchar, integer);"
+ ActiveRecord::Base.connection.execute "DROP view permission_graph_edges;"
+
+ ActiveRecord::Base.connection.execute(%{
+CREATE MATERIALIZED VIEW materialized_permission_view AS
+ WITH RECURSIVE perm_value(name, val) AS (
+ VALUES ('can_read'::text,(1)::smallint), ('can_login'::text,1), ('can_write'::text,2), ('can_manage'::text,3)
+ ), perm_edges(tail_uuid, head_uuid, val, follow, trashed) AS (
+ SELECT links.tail_uuid,
+ links.head_uuid,
+ pv.val,
+ ((pv.val = 3) OR (groups.uuid IS NOT NULL)) AS follow,
+ (0)::smallint AS trashed,
+ (0)::smallint AS followtrash
+ FROM ((public.links
+ LEFT JOIN perm_value pv ON ((pv.name = (links.name)::text)))
+ LEFT JOIN public.groups ON (((pv.val < 3) AND ((groups.uuid)::text = (links.head_uuid)::text))))
+ WHERE ((links.link_class)::text = 'permission'::text)
+ UNION ALL
+ SELECT groups.owner_uuid,
+ groups.uuid,
+ 3,
+ true AS bool,
+ CASE
+ WHEN ((groups.trash_at IS NOT NULL) AND (groups.trash_at < clock_timestamp())) THEN 1
+ ELSE 0
+ END AS "case",
+ 1
+ FROM public.groups
+ ), perm(val, follow, user_uuid, target_uuid, trashed) AS (
+ SELECT (3)::smallint AS val,
+ true AS follow,
+ (users.uuid)::character varying(32) AS user_uuid,
+ (users.uuid)::character varying(32) AS target_uuid,
+ (0)::smallint AS trashed
+ FROM public.users
+ UNION
+ SELECT (LEAST((perm_1.val)::integer, edges.val))::smallint AS val,
+ edges.follow,
+ perm_1.user_uuid,
+ (edges.head_uuid)::character varying(32) AS target_uuid,
+ ((GREATEST((perm_1.trashed)::integer, edges.trashed) * edges.followtrash))::smallint AS trashed
+ FROM (perm perm_1
+ JOIN perm_edges edges ON ((perm_1.follow AND ((edges.tail_uuid)::text = (perm_1.target_uuid)::text))))
+ )
+ SELECT perm.user_uuid,
+ perm.target_uuid,
+ max(perm.val) AS perm_level,
+ CASE perm.follow
+ WHEN true THEN perm.target_uuid
+ ELSE NULL::character varying
+ END AS target_owner_uuid,
+ max(perm.trashed) AS trashed
+ FROM perm
+ GROUP BY perm.user_uuid, perm.target_uuid,
+ CASE perm.follow
+ WHEN true THEN perm.target_uuid
+ ELSE NULL::character varying
+ END
+ WITH NO DATA;
+}
+ )
+
+ add_index :materialized_permission_view, [:trashed, :target_uuid], name: 'permission_target_trashed'
+ add_index :materialized_permission_view, [:user_uuid, :trashed, :perm_level], name: 'permission_target_user_trashed_level'
+ create_table :permission_refresh_lock
+
+ ActiveRecord::Base.connection.execute 'REFRESH MATERIALIZED VIEW materialized_permission_view;'
+ end
+end
-- COMMENT ON EXTENSION pg_trgm IS 'text similarity measurement and index searching based on trigrams';
+--
+-- Name: compute_permission_subgraph(character varying, character varying, integer, character varying); Type: FUNCTION; Schema: public; Owner: -
+--
+
+CREATE FUNCTION public.compute_permission_subgraph(perm_origin_uuid character varying, starting_uuid character varying, starting_perm integer, perm_edge_id character varying) RETURNS TABLE(user_uuid character varying, target_uuid character varying, val integer, traverse_owned boolean)
+ LANGUAGE sql STABLE
+ AS $$
+
+/* The purpose of this function is to compute the permissions for a
+ subgraph of the database, starting from a given edge. The newly
+ computed permissions are used to add and remove rows from the main
+ permissions table.
+
+ perm_origin_uuid: The object that 'gets' the permission.
+
+ starting_uuid: The starting object the permission applies to.
+
+ starting_perm: The permission that perm_origin_uuid 'has' on
+ starting_uuid One of 1, 2, 3 for can_read,
+ can_write, can_manage respectively, or 0 to revoke
+ permissions.
+
+ perm_edge_id: Identifies the permission edge that is being updated.
+ Changes of ownership, this is starting_uuid.
+ For links, this is the uuid of the link object.
+ This is used to override the edge value in the database
+ with starting_perm. This is necessary when revoking
+ permissions because the update happens before edge is
+ actually removed.
+*/
+with
+ /* Starting from starting_uuid, determine the set of objects that
+ could be affected by this permission change.
+
+ Note: We don't traverse users unless it is an "identity"
+ permission (permission origin is self).
+ */
+ perm_from_start(perm_origin_uuid, target_uuid, val, traverse_owned) as (
+
+WITH RECURSIVE
+ traverse_graph(origin_uuid, target_uuid, val, traverse_owned, starting_set) as (
+
+ values (perm_origin_uuid, starting_uuid, starting_perm,
+ should_traverse_owned(starting_uuid, starting_perm),
+ (perm_origin_uuid = starting_uuid or starting_uuid not like '_____-tpzed-_______________'))
+
+ union
+ (select traverse_graph.origin_uuid,
+ edges.head_uuid,
+ least(
+case (edges.edge_id = perm_edge_id)
+ when true then starting_perm
+ else edges.val
+ end
+,
+ traverse_graph.val),
+ should_traverse_owned(edges.head_uuid, edges.val),
+ false
+ from permission_graph_edges as edges, traverse_graph
+ where traverse_graph.target_uuid = edges.tail_uuid
+ and (edges.tail_uuid like '_____-j7d0g-_______________' or
+ traverse_graph.starting_set)))
+ select traverse_graph.origin_uuid, target_uuid, max(val) as val, bool_or(traverse_owned) as traverse_owned from traverse_graph
+ group by (traverse_graph.origin_uuid, target_uuid)
+),
+
+ /* Find other inbound edges that grant permissions to 'targets' in
+ perm_from_start, and compute permissions that originate from
+ those.
+
+ This is necessary for two reasons:
+
+ 1) Other users may have access to a subset of the objects
+ through other permission links than the one we started from.
+ If we don't recompute them, their permission will get dropped.
+
+ 2) There may be more than one path through which a user gets
+ permission to an object. For example, a user owns a project
+ and also shares it can_read with a group the user belongs
+ to. adding the can_read link must not overwrite the existing
+ can_manage permission granted by ownership.
+ */
+ additional_perms(perm_origin_uuid, target_uuid, val, traverse_owned) as (
+
+WITH RECURSIVE
+ traverse_graph(origin_uuid, target_uuid, val, traverse_owned, starting_set) as (
+
+ select edges.tail_uuid as origin_uuid, edges.head_uuid as target_uuid, edges.val,
+ should_traverse_owned(edges.head_uuid, edges.val),
+ edges.head_uuid like '_____-j7d0g-_______________'
+ from permission_graph_edges as edges
+ where edges.edge_id != perm_edge_id and
+ edges.tail_uuid not in (select target_uuid from perm_from_start where target_uuid like '_____-j7d0g-_______________') and
+ edges.head_uuid in (select target_uuid from perm_from_start)
+
+ union
+ (select traverse_graph.origin_uuid,
+ edges.head_uuid,
+ least(
+case (edges.edge_id = perm_edge_id)
+ when true then starting_perm
+ else edges.val
+ end
+,
+ traverse_graph.val),
+ should_traverse_owned(edges.head_uuid, edges.val),
+ false
+ from permission_graph_edges as edges, traverse_graph
+ where traverse_graph.target_uuid = edges.tail_uuid
+ and (edges.tail_uuid like '_____-j7d0g-_______________' or
+ traverse_graph.starting_set)))
+ select traverse_graph.origin_uuid, target_uuid, max(val) as val, bool_or(traverse_owned) as traverse_owned from traverse_graph
+ group by (traverse_graph.origin_uuid, target_uuid)
+),
+
+ /* Combine the permissions computed in the first two phases. */
+ all_perms(perm_origin_uuid, target_uuid, val, traverse_owned) as (
+ select * from perm_from_start
+ union all
+ select * from additional_perms
+ )
+
+ /* The actual query that produces rows to be added or removed
+ from the materialized_permissions table. This is the clever
+ bit.
+
+ Key insights:
+
+ * For every group, the materialized_permissions lists all users
+ that can access to that group.
+
+ * The all_perms subquery has computed permissions on on a set of
+ objects for all inbound "origins", which are users or groups.
+
+ * Permissions through groups are transitive.
+
+ We can infer:
+
+ 1) The materialized_permissions table declares that user X has permission N on group Y
+ 2) The all_perms result has determined group Y has permission M on object Z
+ 3) Therefore, user X has permission min(N, M) on object Z
+
+ This allows us to efficiently determine the set of users that
+ have permissions on the subset of objects, without having to
+ follow the chain of permission back up to find those users.
+
+ In addition, because users always have permission on themselves, this
+ query also makes sure those permission rows are always
+ returned.
+ */
+ select v.user_uuid, v.target_uuid, max(v.perm_level), bool_or(v.traverse_owned) from
+ (select m.user_uuid,
+ u.target_uuid,
+ least(u.val, m.perm_level) as perm_level,
+ u.traverse_owned
+ from all_perms as u, materialized_permissions as m
+ where u.perm_origin_uuid = m.target_uuid AND m.traverse_owned
+ AND (m.user_uuid = m.target_uuid or m.target_uuid not like '_____-tpzed-_______________')
+ union all
+ select target_uuid as user_uuid, target_uuid, 3, true
+ from all_perms
+ where all_perms.target_uuid like '_____-tpzed-_______________') as v
+ group by v.user_uuid, v.target_uuid
+$$;
+
+
+--
+-- Name: project_subtree_with_trash_at(character varying, timestamp without time zone); Type: FUNCTION; Schema: public; Owner: -
+--
+
+CREATE FUNCTION public.project_subtree_with_trash_at(starting_uuid character varying, starting_trash_at timestamp without time zone) RETURNS TABLE(target_uuid character varying, trash_at timestamp without time zone)
+ LANGUAGE sql STABLE
+ AS $$
+/* Starting from a project, recursively traverse all the projects
+ underneath it and return a set of project uuids and trash_at times
+ (may be null). The initial trash_at can be a timestamp or null.
+ The trash_at time propagates downward to groups it owns, i.e. when a
+ group is trashed, everything underneath it in the ownership
+ hierarchy is also considered trashed. However, this is fact is
+ recorded in the trashed_groups table, not by updating trash_at field
+ in the groups table.
+*/
+WITH RECURSIVE
+ project_subtree(uuid, trash_at) as (
+ values (starting_uuid, starting_trash_at)
+ union
+ select groups.uuid, LEAST(project_subtree.trash_at, groups.trash_at)
+ from groups join project_subtree on (groups.owner_uuid = project_subtree.uuid)
+ )
+ select uuid, trash_at from project_subtree;
+$$;
+
+
+--
+-- Name: should_traverse_owned(character varying, integer); Type: FUNCTION; Schema: public; Owner: -
+--
+
+CREATE FUNCTION public.should_traverse_owned(starting_uuid character varying, starting_perm integer) RETURNS boolean
+ LANGUAGE sql IMMUTABLE
+ AS $$
+/* Helper function. Determines if permission on an object implies
+ transitive permission to things the object owns. This is always
+ true for groups, but only true for users when the permission level
+ is can_manage.
+*/
+select starting_uuid like '_____-j7d0g-_______________' or
+ (starting_uuid like '_____-tpzed-_______________' and starting_perm >= 3);
+$$;
+
+
SET default_tablespace = '';
SET default_with_oids = false;
--
--- Name: users; Type: TABLE; Schema: public; Owner: -
+-- Name: materialized_permissions; Type: TABLE; Schema: public; Owner: -
--
-CREATE TABLE public.users (
- id integer NOT NULL,
- uuid character varying(255),
- owner_uuid character varying(255) NOT NULL,
- created_at timestamp without time zone NOT NULL,
- modified_by_client_uuid character varying(255),
- modified_by_user_uuid character varying(255),
- modified_at timestamp without time zone,
- email character varying(255),
- first_name character varying(255),
- last_name character varying(255),
- identity_url character varying(255),
- is_admin boolean,
- prefs text,
- updated_at timestamp without time zone NOT NULL,
- default_owner_uuid character varying(255),
- is_active boolean DEFAULT false,
- username character varying(255),
- redirect_to_user_uuid character varying
+CREATE TABLE public.materialized_permissions (
+ user_uuid character varying,
+ target_uuid character varying,
+ perm_level integer,
+ traverse_owned boolean
);
---
--- Name: materialized_permission_view; Type: MATERIALIZED VIEW; Schema: public; Owner: -
---
-
-CREATE MATERIALIZED VIEW public.materialized_permission_view AS
- WITH RECURSIVE perm_value(name, val) AS (
- VALUES ('can_read'::text,(1)::smallint), ('can_login'::text,1), ('can_write'::text,2), ('can_manage'::text,3)
- ), perm_edges(tail_uuid, head_uuid, val, follow, trashed) AS (
- SELECT links.tail_uuid,
- links.head_uuid,
- pv.val,
- ((pv.val = 3) OR (groups.uuid IS NOT NULL)) AS follow,
- (0)::smallint AS trashed,
- (0)::smallint AS followtrash
- FROM ((public.links
- LEFT JOIN perm_value pv ON ((pv.name = (links.name)::text)))
- LEFT JOIN public.groups ON (((pv.val < 3) AND ((groups.uuid)::text = (links.head_uuid)::text))))
- WHERE ((links.link_class)::text = 'permission'::text)
- UNION ALL
- SELECT groups.owner_uuid,
- groups.uuid,
- 3,
- true AS bool,
- CASE
- WHEN ((groups.trash_at IS NOT NULL) AND (groups.trash_at < clock_timestamp())) THEN 1
- ELSE 0
- END AS "case",
- 1
- FROM public.groups
- ), perm(val, follow, user_uuid, target_uuid, trashed) AS (
- SELECT (3)::smallint AS val,
- true AS follow,
- (users.uuid)::character varying(32) AS user_uuid,
- (users.uuid)::character varying(32) AS target_uuid,
- (0)::smallint AS trashed
- FROM public.users
- UNION
- SELECT (LEAST((perm_1.val)::integer, edges.val))::smallint AS val,
- edges.follow,
- perm_1.user_uuid,
- (edges.head_uuid)::character varying(32) AS target_uuid,
- ((GREATEST((perm_1.trashed)::integer, edges.trashed) * edges.followtrash))::smallint AS trashed
- FROM (perm perm_1
- JOIN perm_edges edges ON ((perm_1.follow AND ((edges.tail_uuid)::text = (perm_1.target_uuid)::text))))
- )
- SELECT perm.user_uuid,
- perm.target_uuid,
- max(perm.val) AS perm_level,
- CASE perm.follow
- WHEN true THEN perm.target_uuid
- ELSE NULL::character varying
- END AS target_owner_uuid,
- max(perm.trashed) AS trashed
- FROM perm
- GROUP BY perm.user_uuid, perm.target_uuid,
- CASE perm.follow
- WHEN true THEN perm.target_uuid
- ELSE NULL::character varying
- END
- WITH NO DATA;
-
-
--
-- Name: nodes; Type: TABLE; Schema: public; Owner: -
--
--
--- Name: permission_refresh_lock; Type: TABLE; Schema: public; Owner: -
+-- Name: users; Type: TABLE; Schema: public; Owner: -
--
-CREATE TABLE public.permission_refresh_lock (
- id integer NOT NULL
+CREATE TABLE public.users (
+ id integer NOT NULL,
+ uuid character varying(255),
+ owner_uuid character varying(255) NOT NULL,
+ created_at timestamp without time zone NOT NULL,
+ modified_by_client_uuid character varying(255),
+ modified_by_user_uuid character varying(255),
+ modified_at timestamp without time zone,
+ email character varying(255),
+ first_name character varying(255),
+ last_name character varying(255),
+ identity_url character varying(255),
+ is_admin boolean,
+ prefs text,
+ updated_at timestamp without time zone NOT NULL,
+ default_owner_uuid character varying(255),
+ is_active boolean DEFAULT false,
+ username character varying(255),
+ redirect_to_user_uuid character varying
);
--
--- Name: permission_refresh_lock_id_seq; Type: SEQUENCE; Schema: public; Owner: -
---
-
-CREATE SEQUENCE public.permission_refresh_lock_id_seq
- START WITH 1
- INCREMENT BY 1
- NO MINVALUE
- NO MAXVALUE
- CACHE 1;
-
-
---
--- Name: permission_refresh_lock_id_seq; Type: SEQUENCE OWNED BY; Schema: public; Owner: -
---
-
-ALTER SEQUENCE public.permission_refresh_lock_id_seq OWNED BY public.permission_refresh_lock.id;
+-- Name: permission_graph_edges; Type: VIEW; Schema: public; Owner: -
+--
+
+CREATE VIEW public.permission_graph_edges AS
+ SELECT groups.owner_uuid AS tail_uuid,
+ groups.uuid AS head_uuid,
+ 3 AS val,
+ groups.uuid AS edge_id
+ FROM public.groups
+UNION ALL
+ SELECT users.owner_uuid AS tail_uuid,
+ users.uuid AS head_uuid,
+ 3 AS val,
+ users.uuid AS edge_id
+ FROM public.users
+UNION ALL
+ SELECT users.uuid AS tail_uuid,
+ users.uuid AS head_uuid,
+ 3 AS val,
+ ''::character varying AS edge_id
+ FROM public.users
+UNION ALL
+ SELECT links.tail_uuid,
+ links.head_uuid,
+ CASE
+ WHEN ((links.name)::text = 'can_read'::text) THEN 1
+ WHEN ((links.name)::text = 'can_login'::text) THEN 1
+ WHEN ((links.name)::text = 'can_write'::text) THEN 2
+ WHEN ((links.name)::text = 'can_manage'::text) THEN 3
+ ELSE 0
+ END AS val,
+ links.uuid AS edge_id
+ FROM public.links
+ WHERE ((links.link_class)::text = 'permission'::text);
--
ALTER SEQUENCE public.traits_id_seq OWNED BY public.traits.id;
+--
+-- Name: trashed_groups; Type: TABLE; Schema: public; Owner: -
+--
+
+CREATE TABLE public.trashed_groups (
+ group_uuid character varying,
+ trash_at timestamp without time zone
+);
+
+
--
-- Name: users_id_seq; Type: SEQUENCE; Schema: public; Owner: -
--
ALTER TABLE ONLY public.nodes ALTER COLUMN id SET DEFAULT nextval('public.nodes_id_seq'::regclass);
---
--- Name: permission_refresh_lock id; Type: DEFAULT; Schema: public; Owner: -
---
-
-ALTER TABLE ONLY public.permission_refresh_lock ALTER COLUMN id SET DEFAULT nextval('public.permission_refresh_lock_id_seq'::regclass);
-
-
--
-- Name: pipeline_instances id; Type: DEFAULT; Schema: public; Owner: -
--
ADD CONSTRAINT nodes_pkey PRIMARY KEY (id);
---
--- Name: permission_refresh_lock permission_refresh_lock_pkey; Type: CONSTRAINT; Schema: public; Owner: -
---
-
-ALTER TABLE ONLY public.permission_refresh_lock
- ADD CONSTRAINT permission_refresh_lock_pkey PRIMARY KEY (id);
-
-
--
-- Name: pipeline_instances pipeline_instances_pkey; Type: CONSTRAINT; Schema: public; Owner: -
--
CREATE UNIQUE INDEX index_traits_on_uuid ON public.traits USING btree (uuid);
+--
+-- Name: index_trashed_groups_on_group_uuid; Type: INDEX; Schema: public; Owner: -
+--
+
+CREATE UNIQUE INDEX index_trashed_groups_on_group_uuid ON public.trashed_groups USING btree (group_uuid);
+
+
--
-- Name: index_users_on_created_at; Type: INDEX; Schema: public; Owner: -
--
--
--- Name: permission_target_trashed; Type: INDEX; Schema: public; Owner: -
+-- Name: permission_target; Type: INDEX; Schema: public; Owner: -
--
-CREATE INDEX permission_target_trashed ON public.materialized_permission_view USING btree (trashed, target_uuid);
+CREATE INDEX permission_target ON public.materialized_permissions USING btree (target_uuid);
--
--- Name: permission_target_user_trashed_level; Type: INDEX; Schema: public; Owner: -
+-- Name: permission_user_target; Type: INDEX; Schema: public; Owner: -
--
-CREATE INDEX permission_target_user_trashed_level ON public.materialized_permission_view USING btree (user_uuid, trashed, perm_level);
+CREATE UNIQUE INDEX permission_user_target ON public.materialized_permissions USING btree (user_uuid, target_uuid);
--
('20190523180148'),
('20190808145904'),
('20190809135453'),
-('20190905151603');
+('20190905151603'),
+('20200501150153');
--- /dev/null
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+# These constants are used in both
+# db/migrate/20200501150153_permission_table and update_permissions
+#
+# This file allows them to be easily imported by both to avoid duplication.
+#
+# Don't mess with this! Any changes will affect both the current
+# update_permissions and the past migration. If you are tinkering
+# with the permission system and need to change how
+# PERM_QUERY_TEMPLATE, refresh_trashed or refresh_permissions works,
+# you should make a new file with your modified functions and have
+# update_permissions reference that file instead.
+
+PERMISSION_VIEW = "materialized_permissions"
+
+TRASHED_GROUPS = "trashed_groups"
+
+# We need to use this parameterized query in a few different places,
+# including as a subquery in a larger query.
+#
+# There's basically two options, the way I did this originally was to
+# put this in a postgres function and do a lateral join over it.
+# However, postgres functions impose an optimization barrier, and
+# possibly have other overhead with temporary tables, so I ended up
+# going with the brute force approach of inlining the whole thing.
+#
+# The two substitutions are "base_case" which determines the initial
+# set of permission origins and "edge_perm" which is used to ensure
+# that the new permission takes precedence over the one in the edges
+# table (but some queries don't need that.)
+#
+PERM_QUERY_TEMPLATE = %{
+WITH RECURSIVE
+ traverse_graph(origin_uuid, target_uuid, val, traverse_owned, starting_set) as (
+ %{base_case}
+ union
+ (select traverse_graph.origin_uuid,
+ edges.head_uuid,
+ least(%{edge_perm},
+ traverse_graph.val),
+ should_traverse_owned(edges.head_uuid, edges.val),
+ false
+ from permission_graph_edges as edges, traverse_graph
+ where traverse_graph.target_uuid = edges.tail_uuid
+ and (edges.tail_uuid like '_____-j7d0g-_______________' or
+ traverse_graph.starting_set)))
+ select traverse_graph.origin_uuid, target_uuid, max(val) as val, bool_or(traverse_owned) as traverse_owned from traverse_graph
+ group by (traverse_graph.origin_uuid, target_uuid)
+}
+
+def refresh_trashed
+ ActiveRecord::Base.transaction do
+ ActiveRecord::Base.connection.execute("LOCK TABLE #{TRASHED_GROUPS}")
+ ActiveRecord::Base.connection.execute("DELETE FROM #{TRASHED_GROUPS}")
+
+ # Helper populate trashed_groups table. This starts with
+ # each group owned by a user and computes the subtree under that
+ # group to find any groups that are trashed.
+ ActiveRecord::Base.connection.execute(%{
+INSERT INTO #{TRASHED_GROUPS}
+select ps.target_uuid as group_uuid, ps.trash_at from groups,
+ lateral project_subtree_with_trash_at(groups.uuid, groups.trash_at) ps
+ where groups.owner_uuid like '_____-tpzed-_______________'
+})
+ end
+end
+
+def refresh_permissions
+ ActiveRecord::Base.transaction do
+ ActiveRecord::Base.connection.execute("LOCK TABLE #{PERMISSION_VIEW}")
+ ActiveRecord::Base.connection.execute("DELETE FROM #{PERMISSION_VIEW}")
+
+ ActiveRecord::Base.connection.execute %{
+INSERT INTO materialized_permissions
+ #{PERM_QUERY_TEMPLATE % {:base_case => %{
+ select uuid, uuid, 3, true, true from users
+},
+:edge_perm => 'edges.val'
+} }
+}, "refresh_permission_view.do"
+ end
+end
+++ /dev/null
-# Copyright (C) The Arvados Authors. All rights reserved.
-#
-# SPDX-License-Identifier: AGPL-3.0
-
-PERMISSION_VIEW = "materialized_permission_view"
-
-def do_refresh_permission_view
- ActiveRecord::Base.transaction do
- ActiveRecord::Base.connection.execute("LOCK TABLE permission_refresh_lock")
- ActiveRecord::Base.connection.execute("REFRESH MATERIALIZED VIEW #{PERMISSION_VIEW}")
- end
-end
-
-def refresh_permission_view(async=false)
- if async and Rails.configuration.API.AsyncPermissionsUpdateInterval > 0
- exp = Rails.configuration.API.AsyncPermissionsUpdateInterval.seconds
- need = false
- Rails.cache.fetch('AsyncRefreshPermissionView', expires_in: exp) do
- need = true
- end
- if need
- # Schedule a new permission update and return immediately
- Thread.new do
- Thread.current.abort_on_exception = false
- begin
- sleep(exp)
- Rails.cache.delete('AsyncRefreshPermissionView')
- do_refresh_permission_view
- rescue => e
- Rails.logger.error "Updating permission view: #{e}\n#{e.backtrace.join("\n\t")}"
- ensure
- ActiveRecord::Base.connection.close
- end
- end
- true
- end
- else
- do_refresh_permission_view
- end
-end
--- /dev/null
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+require '20200501150153_permission_table_constants'
+
+REVOKE_PERM = 0
+CAN_MANAGE_PERM = 3
+
+def update_permissions perm_origin_uuid, starting_uuid, perm_level, edge_id=nil
+ #
+ # Update a subset of the permission table affected by adding or
+ # removing a particular permission relationship (ownership or a
+ # permission link).
+ #
+ # perm_origin_uuid: This is the object that 'gets' the permission.
+ # It is the owner_uuid or tail_uuid.
+ #
+ # starting_uuid: The object we are computing permission for (or head_uuid)
+ #
+ # perm_level: The level of permission that perm_origin_uuid gets for starting_uuid.
+ #
+ # perm_level is a number from 0-3
+ # can_read=1
+ # can_write=2
+ # can_manage=3
+ # or call with perm_level=0 to revoke permissions
+ #
+ # check: for testing/debugging, compare the result of the
+ # incremental update against a full table recompute. Throws an
+ # error if the contents are not identical (ie they produce different
+ # permission results)
+
+ # Theory of operation
+ #
+ # Give a change in a specific permission relationship, we recompute
+ # the set of permissions (for all users) that could possibly be
+ # affected by that relationship. For example, if a project is
+ # shared with another user, we recompute all permissions for all
+ # projects in the hierarchy. This returns a set of updated
+ # permissions, which we stash in a temporary table.
+ #
+ # Then, for each user_uuid/target_uuid in the updated permissions
+ # result set we insert/update a permission row in
+ # materialized_permissions, and delete any rows that exist in
+ # materialized_permissions that are not in the result set or have
+ # perm_level=0.
+ #
+ # see db/migrate/20200501150153_permission_table.rb for details on
+ # how the permissions are computed.
+
+ if edge_id.nil?
+ # For changes of ownership, edge_id is starting_uuid. In turns
+ # out most invocations of update_permissions are for changes of
+ # ownership, so make this parameter optional to reduce
+ # clutter.
+ # For permission links, the uuid of the link object will be passed in for edge_id.
+ edge_id = starting_uuid
+ end
+
+ ActiveRecord::Base.transaction do
+
+ # "Conflicts with the ROW EXCLUSIVE, SHARE UPDATE EXCLUSIVE, SHARE
+ # ROW EXCLUSIVE, EXCLUSIVE, and ACCESS EXCLUSIVE lock modes. This
+ # mode protects a table against concurrent data changes."
+ ActiveRecord::Base.connection.execute "LOCK TABLE #{PERMISSION_VIEW} in SHARE MODE"
+
+ # Workaround for
+ # BUG #15160: planner overestimates number of rows in join when there are more than 200 rows coming from CTE
+ # https://www.postgresql.org/message-id/152395805004.19366.3107109716821067806@wrigleys.postgresql.org
+ #
+ # For a crucial join in the compute_permission_subgraph() query, the
+ # planner mis-estimates the number of rows in a Common Table
+ # Expression (CTE, this is a subquery in a WITH clause) and as a
+ # result it chooses the wrong join order. The join starts with the
+ # permissions table because it mistakenly thinks
+ # count(materalized_permissions) < count(new computed permissions)
+ # when actually it is the other way around.
+ #
+ # Because of the incorrect join order, it choose the wrong join
+ # strategy (merge join, which works best when two tables are roughly
+ # the same size). As a workaround, we can tell it not to use that
+ # join strategy, this causes it to pick hash join instead, which
+ # turns out to be a bit better. However, because the join order is
+ # still wrong, we don't get the full benefit of the index.
+ #
+ # This is very unfortunate because it makes the query performance
+ # dependent on the size of the materalized_permissions table, when
+ # the goal of this design was to make permission updates scale-free
+ # and only depend on the number of permissions affected and not the
+ # total table size. In several hours of researching I wasn't able
+ # to find a way to force the correct join order, so I'm calling it
+ # here and I have to move on.
+ #
+ # This is apparently addressed in Postgres 12, but I developed &
+ # tested this on Postgres 9.6, so in the future we should reevaluate
+ # the performance & query plan on Postgres 12.
+ #
+ # https://git.furworks.de/opensourcemirror/postgresql/commit/a314c34079cf06d05265623dd7c056f8fa9d577f
+ #
+ # Disable merge join for just this query (also local for this transaction), then reenable it.
+ ActiveRecord::Base.connection.exec_query "SET LOCAL enable_mergejoin to false;"
+
+ temptable_perms = "temp_perms_#{rand(2**64).to_s(10)}"
+ ActiveRecord::Base.connection.exec_query %{
+create temporary table #{temptable_perms} on commit drop
+as select * from compute_permission_subgraph($1, $2, $3, $4)
+},
+ 'update_permissions.select',
+ [[nil, perm_origin_uuid],
+ [nil, starting_uuid],
+ [nil, perm_level],
+ [nil, edge_id]]
+
+ ActiveRecord::Base.connection.exec_query "SET LOCAL enable_mergejoin to true;"
+
+ ActiveRecord::Base.connection.exec_delete %{
+delete from #{PERMISSION_VIEW} where
+ target_uuid in (select target_uuid from #{temptable_perms}) and
+ not exists (select 1 from #{temptable_perms}
+ where target_uuid=#{PERMISSION_VIEW}.target_uuid and
+ user_uuid=#{PERMISSION_VIEW}.user_uuid and
+ val>0)
+},
+ "update_permissions.delete"
+
+ ActiveRecord::Base.connection.exec_query %{
+insert into #{PERMISSION_VIEW} (user_uuid, target_uuid, perm_level, traverse_owned)
+ select user_uuid, target_uuid, val as perm_level, traverse_owned from #{temptable_perms} where val>0
+on conflict (user_uuid, target_uuid) do update set perm_level=EXCLUDED.perm_level, traverse_owned=EXCLUDED.traverse_owned;
+},
+ "update_permissions.insert"
+
+ if perm_level>0
+ check_permissions_against_full_refresh
+ end
+ end
+end
+
+
+def check_permissions_against_full_refresh
+ # No-op except when running tests
+ return unless Rails.env == 'test' and !Thread.current[:no_check_permissions_against_full_refresh]
+
+ # For checking correctness of the incremental permission updates.
+ # Check contents of the current 'materialized_permission' table
+ # against a from-scratch permission refresh.
+
+ q1 = ActiveRecord::Base.connection.exec_query %{
+select user_uuid, target_uuid, perm_level, traverse_owned from #{PERMISSION_VIEW}
+order by user_uuid, target_uuid
+}, "check_permissions_against_full_refresh.permission_table"
+
+ q2 = ActiveRecord::Base.connection.exec_query %{
+ select pq.origin_uuid as user_uuid, target_uuid, pq.val as perm_level, pq.traverse_owned from (
+ #{PERM_QUERY_TEMPLATE % {:base_case => %{
+ select uuid, uuid, 3, true, true from users
+},
+:edge_perm => 'edges.val'
+} }) as pq order by origin_uuid, target_uuid
+}, "check_permissions_against_full_refresh.full_recompute"
+
+ if q1.count != q2.count
+ puts "Didn't match incremental+: #{q1.count} != full refresh-: #{q2.count}"
+ end
+
+ if q1.count > q2.count
+ q1.each_with_index do |r, i|
+ if r != q2[i]
+ puts "+#{r}\n-#{q2[i]}"
+ raise "Didn't match"
+ end
+ end
+ else
+ q2.each_with_index do |r, i|
+ if r != q1[i]
+ puts "+#{q1[i]}\n-#{r}"
+ raise "Didn't match"
+ end
+ end
+ end
+end
+
+def skip_check_permissions_against_full_refresh
+ check_perm_was = Thread.current[:no_check_permissions_against_full_refresh]
+ Thread.current[:no_check_permissions_against_full_refresh] = true
+ begin
+ yield
+ ensure
+ Thread.current[:no_check_permissions_against_full_refresh] = check_perm_was
+ end
+end
+
+# Used to account for permissions that a user gains by having
+# can_manage on another user.
+#
+# note: in theory a user could have can_manage access to a user
+# through multiple levels, that isn't handled here (would require a
+# recursive query). I think that's okay because users getting
+# transitive access through "can_manage" on a user is is rarely/never
+# used feature and something we probably want to deprecate and remove.
+USER_UUIDS_SUBQUERY_TEMPLATE = %{
+select target_uuid from materialized_permissions where user_uuid in (%{user})
+and target_uuid like '_____-tpzed-_______________' and traverse_owned=true and perm_level >= %{perm_level}
+}
uuid: zzzzz-j7d0g-48foin4vonvc2at
owner_uuid: zzzzz-tpzed-000000000000000
name: Administrators of a subset of users
+ group_class: role
aproject:
uuid: zzzzz-j7d0g-v955i6s2oi1cbso
uuid: zzzzz-j7d0g-ptt1ou6a9lxrv07
owner_uuid: zzzzz-tpzed-d9tiejq69daie8f
name: Active user has can_manage
+ group_class: project
# Group for testing granting permission between users who share a group.
group_for_sharing_tests:
trash_at: 2001-01-01T00:00:00Z
delete_at: 2038-03-01T00:00:00Z
is_trashed: false
- modified_at: 2001-01-01T00:00:00Z
\ No newline at end of file
+ modified_at: 2001-01-01T00:00:00Z
organization: example.com
role: Computational biologist
getting_started_shown: 2015-03-26 12:34:56.789000000 Z
+
+has_can_login_permission:
+ owner_uuid: zzzzz-tpzed-000000000000000
+ uuid: zzzzz-tpzed-xabcdjxw79nv3jz
+ email: can-login-user@arvados.local
+ modified_by_client_uuid: zzzzz-ozdt8-teyxzyd8qllg11h
+ modified_by_user_uuid: zzzzz-tpzed-xurymjxw79nv3jz
+ first_name: Can_login
+ last_name: User
+ identity_url: https://can-login-user.openid.local
+ is_active: true
+ is_admin: false
+ modified_at: 2015-03-26 12:34:56.789000000 Z
+ username: can-login-user
### trashed project tests ###
- [:active, :admin].each do |auth|
+ #
+ # The structure is
+ #
+ # trashed_project (zzzzz-j7d0g-trashedproject1)
+ # trashed_subproject (zzzzz-j7d0g-trashedproject2)
+ # trashed_subproject3 (zzzzz-j7d0g-trashedproject3)
+ # zzzzz-xvhdp-cr5trashedcontr
+
+ [:active,
+ :admin].each do |auth|
# project: to query, to untrash, is visible, parent contents listing success
- [[:trashed_project, [], false, true],
+ [
+ [:trashed_project, [], false, true],
[:trashed_project, [:trashed_project], true, true],
[:trashed_subproject, [], false, false],
[:trashed_subproject, [:trashed_project], true, true],
assert_response :success
end
- test "create request with async=true defers permissions update" do
+ test "create request with async=true does not defer permissions update" do
Rails.configuration.API.AsyncPermissionsUpdateInterval = 1 # second
name = "Random group #{rand(1000)}"
assert_equal nil, Group.find_by_name(name)
+ # Following the implementation of incremental permission updates
+ # (#16007) the async flag is now a no-op. Permission changes are
+ # visible immediately.
+
# Trigger the asynchronous permission update by using async=true parameter.
post "/arvados/v1/groups",
params: {
headers: auth(:active)
assert_response 202
- # The group exists on the database, but it's not accessible yet.
+ # The group exists in the database
assert_not_nil Group.find_by_name(name)
get "/arvados/v1/groups",
params: {
},
headers: auth(:active)
assert_response 200
- assert_equal 0, json_response['items_available']
+ assert_equal 1, json_response['items_available']
# Wait a bit and try again.
sleep(1)
end
end
end
- User.invalidate_permissions_cache
+ refresh_permissions
end
end)
end
#
# SPDX-License-Identifier: AGPL-3.0
+require 'update_permissions'
+
ENV["RAILS_ENV"] = "test"
unless ENV["NO_COVERAGE_TEST"]
begin
end
# Ensure permissions are computed from the test fixtures.
-User.invalidate_permissions_cache
+refresh_permissions
+refresh_trashed
test "delete referring links in SweepTrashedObjects" do
uuid = collections(:trashed_on_next_sweep).uuid
act_as_system_user do
+ assert_raises ActiveRecord::RecordInvalid do
+ # Cannot create because :trashed_on_next_sweep is already trashed
+ Link.create!(head_uuid: uuid,
+ tail_uuid: system_user_uuid,
+ link_class: 'whatever',
+ name: 'something')
+ end
+
+ # Bump trash_at to now + 1 minute
+ Collection.where(uuid: uuid).
+ update(trash_at: db_current_time + (1).minute)
+
+ # Not considered trashed now
Link.create!(head_uuid: uuid,
tail_uuid: system_user_uuid,
link_class: 'whatever',
"new #{o_class} should really be in DB")
old_uuid = o.uuid
new_uuid = o.uuid.sub(/..........$/, rand(2**256).to_s(36)[0..9])
- assert(o.update_attributes(uuid: new_uuid),
- "should change #{o_class} uuid from #{old_uuid} to #{new_uuid}")
+ if o.respond_to? :update_uuid
+ o.update_uuid(new_uuid: new_uuid)
+ else
+ assert(o.update_attributes(uuid: new_uuid),
+ "should change #{o_class} uuid from #{old_uuid} to #{new_uuid}")
+ end
assert_equal(false, o_class.where(uuid: old_uuid).any?,
"#{old_uuid} should disappear when renamed to #{new_uuid}")
end
assert_equal(true, Specimen.where(owner_uuid: o.uuid).any?,
"need something to be owned by #{o.uuid} for this test")
- assert_raises(ActiveRecord::DeleteRestrictionError,
- "should not delete #{ofixt} that owns objects") do
- o.destroy
+ skip_check_permissions_against_full_refresh do
+ assert_raises(ActiveRecord::DeleteRestrictionError,
+ "should not delete #{ofixt} that owns objects") do
+ o.destroy
+ end
end
end
assert User.where(uuid: o.uuid).any?, "new User should really be in DB"
assert_equal(true, o.update_attributes(owner_uuid: o.uuid),
"setting owner to self should work")
- assert(o.destroy, "should delete User that owns self")
+
+ skip_check_permissions_against_full_refresh do
+ assert(o.destroy, "should delete User that owns self")
+ end
+
assert_equal(false, User.where(uuid: o.uuid).any?,
"#{o.uuid} should not be in DB after deleting")
+ check_permissions_against_full_refresh
end
test "change uuid of User that owns self" do
"setting owner to self should work")
old_uuid = o.uuid
new_uuid = o.uuid.sub(/..........$/, rand(2**256).to_s(36)[0..9])
- assert(o.update_attributes(uuid: new_uuid),
- "should change uuid of User that owns self")
+ o.update_uuid(new_uuid: new_uuid)
+ o = User.find_by_uuid(new_uuid)
assert_equal(false, User.where(uuid: old_uuid).any?,
"#{old_uuid} should not be in DB after deleting")
assert_equal(true, User.where(uuid: new_uuid).any?,
test "Grant permissions on an object I own" do
set_user_from_auth :active_trustedclient
- ob = Specimen.create
+ ob = Collection.create
assert ob.save
# Ensure I have permission to manage this group even when its owner changes
test "Delete permission links when deleting an object" do
set_user_from_auth :active_trustedclient
- ob = Specimen.create!
+ ob = Collection.create!
Link.create!(tail_uuid: users(:active).uuid,
head_uuid: ob.uuid,
link_class: 'permission',
test "permission links owned by root" do
set_user_from_auth :active_trustedclient
- ob = Specimen.create!
+ ob = Collection.create!
perm_link = Link.create!(tail_uuid: users(:active).uuid,
head_uuid: ob.uuid,
link_class: 'permission',
test "readable_by" do
set_user_from_auth :active_trustedclient
- ob = Specimen.create!
+ ob = Collection.create!
Link.create!(tail_uuid: users(:active).uuid,
head_uuid: ob.uuid,
link_class: 'permission',
name: 'can_read')
- assert Specimen.readable_by(users(:active)).where(uuid: ob.uuid).any?, "user does not have read permission"
+ assert Collection.readable_by(users(:active)).where(uuid: ob.uuid).any?, "user does not have read permission"
end
test "writable_by" do
set_user_from_auth :active_trustedclient
- ob = Specimen.create!
+ ob = Collection.create!
Link.create!(tail_uuid: users(:active).uuid,
head_uuid: ob.uuid,
link_class: 'permission',
assert ob.writable_by.include?(users(:active).uuid), "user does not have write permission"
end
+ test "update permission link" do
+ set_user_from_auth :admin
+
+ grp = Group.create! name: "blah project", group_class: "project"
+ ob = Collection.create! owner_uuid: grp.uuid
+
+ assert !users(:active).can?(write: ob)
+ assert !users(:active).can?(read: ob)
+
+ l1 = Link.create!(tail_uuid: users(:active).uuid,
+ head_uuid: grp.uuid,
+ link_class: 'permission',
+ name: 'can_write')
+
+ assert users(:active).can?(write: ob)
+ assert users(:active).can?(read: ob)
+
+ l1.update_attributes!(name: 'can_read')
+
+ assert !users(:active).can?(write: ob)
+ assert users(:active).can?(read: ob)
+
+ l1.destroy
+
+ assert !users(:active).can?(write: ob)
+ assert !users(:active).can?(read: ob)
+ end
+
test "writable_by reports requesting user's own uuid for a writable project" do
invited_to_write = users(:project_viewer)
group = groups(:asubproject)
test "user owns group, group can_manage object's group, user can add permissions" do
set_user_from_auth :admin
- owner_grp = Group.create!(owner_uuid: users(:active).uuid)
-
- sp_grp = Group.create!
- sp = Specimen.create!(owner_uuid: sp_grp.uuid)
+ owner_grp = Group.create!(owner_uuid: users(:active).uuid, group_class: "role")
+ sp_grp = Group.create!(group_class: "project")
Link.create!(link_class: 'permission',
name: 'can_manage',
tail_uuid: owner_grp.uuid,
head_uuid: sp_grp.uuid)
+ sp = Collection.create!(owner_uuid: sp_grp.uuid)
+
# active user owns owner_grp, which has can_manage permission on sp_grp
# user should be able to add permissions on sp.
set_user_from_auth :active_trustedclient
skip "can_manage permission on a non-group object" do
set_user_from_auth :admin
- ob = Specimen.create!
+ ob = Collection.create!
# grant can_manage permission to active
perm_link = Link.create!(tail_uuid: users(:active).uuid,
head_uuid: ob.uuid,
test "user without can_manage permission may not modify permission link" do
set_user_from_auth :admin
- ob = Specimen.create!
+ ob = Collection.create!
# grant can_manage permission to active
perm_link = Link.create!(tail_uuid: users(:active).uuid,
head_uuid: ob.uuid,
manager = create :active_user, first_name: "Manage", last_name: "Er"
minion = create :active_user, first_name: "Min", last_name: "Ion"
minions_specimen = act_as_user minion do
- Specimen.create!
+ g = Group.create! name: "minon project", group_class: "project"
+ Collection.create! owner_uuid: g.uuid
end
# Manager creates a group. (Make sure it doesn't magically give
# anyone any additional permissions.)
create(:permission_link,
name: 'can_manage', tail_uuid: manager.uuid, head_uuid: minion.uuid)
end
- assert_empty(Specimen
+ assert_empty(Collection
.readable_by(manager)
.where(uuid: minions_specimen.uuid),
"manager saw the minion's private stuff")
act_as_user manager do
# Now, manager can read and write Minion's stuff.
- assert_not_empty(Specimen
+ assert_not_empty(Collection
.readable_by(manager)
.where(uuid: minions_specimen.uuid),
"manager could not find minion's specimen by uuid")
"#{a.first_name} should be able to see 'b' in the user list")
a_specimen = act_as_user a do
- Specimen.create!
+ Collection.create!
end
- assert_not_empty(Specimen.readable_by(a).where(uuid: a_specimen.uuid),
- "A cannot read own Specimen, following test probably useless.")
- assert_empty(Specimen.readable_by(b).where(uuid: a_specimen.uuid),
- "B can read A's Specimen")
+ assert_not_empty(Collection.readable_by(a).where(uuid: a_specimen.uuid),
+ "A cannot read own Collection, following test probably useless.")
+ assert_empty(Collection.readable_by(b).where(uuid: a_specimen.uuid),
+ "B can read A's Collection")
[a,b].each do |u|
assert_empty(User.readable_by(u).where(uuid: other.uuid),
"#{u.first_name} can see OTHER in the user list")
test "cannot create with owner = unwritable user" do
set_user_from_auth :rominiadmin
assert_raises ArvadosModel::PermissionDeniedError, "created with owner = unwritable user" do
- Specimen.create!(owner_uuid: users(:active).uuid)
+ Collection.create!(owner_uuid: users(:active).uuid)
end
end
test "cannot change owner to unwritable user" do
set_user_from_auth :rominiadmin
- ob = Specimen.create!
+ ob = Collection.create!
assert_raises ArvadosModel::PermissionDeniedError, "changed owner to unwritable user" do
ob.update_attributes!(owner_uuid: users(:active).uuid)
end
test "cannot create with owner = unwritable group" do
set_user_from_auth :rominiadmin
assert_raises ArvadosModel::PermissionDeniedError, "created with owner = unwritable group" do
- Specimen.create!(owner_uuid: groups(:aproject).uuid)
+ Collection.create!(owner_uuid: groups(:aproject).uuid)
end
end
test "cannot change owner to unwritable group" do
set_user_from_auth :rominiadmin
- ob = Specimen.create!
+ ob = Collection.create!
assert_raises ArvadosModel::PermissionDeniedError, "changed owner to unwritable group" do
ob.update_attributes!(owner_uuid: groups(:aproject).uuid)
end
assert_not_empty container_logs(:running_older, :anonymous)
end
+
+ test "add user to group, then remove them" do
+ set_user_from_auth :admin
+ grp = Group.create!(owner_uuid: system_user_uuid, group_class: "role")
+ col = Collection.create!(owner_uuid: grp.uuid)
+ assert_empty Collection.readable_by(users(:active)).where(uuid: col.uuid)
+ assert_empty User.readable_by(users(:active)).where(uuid: users(:project_viewer).uuid)
+
+ l1 = Link.create!(tail_uuid: users(:active).uuid,
+ head_uuid: grp.uuid,
+ link_class: 'permission',
+ name: 'can_read')
+ l2 = Link.create!(tail_uuid: grp.uuid,
+ head_uuid: users(:active).uuid,
+ link_class: 'permission',
+ name: 'can_read')
+
+ l3 = Link.create!(tail_uuid: users(:project_viewer).uuid,
+ head_uuid: grp.uuid,
+ link_class: 'permission',
+ name: 'can_read')
+ l4 = Link.create!(tail_uuid: grp.uuid,
+ head_uuid: users(:project_viewer).uuid,
+ link_class: 'permission',
+ name: 'can_read')
+
+ assert Collection.readable_by(users(:active)).where(uuid: col.uuid).first
+ assert User.readable_by(users(:active)).where(uuid: users(:project_viewer).uuid).first
+
+ l1.destroy
+ l2.destroy
+
+ assert_empty Collection.readable_by(users(:active)).where(uuid: col.uuid)
+ assert_empty User.readable_by(users(:active)).where(uuid: users(:project_viewer).uuid)
+
+ end
+
+
+ test "add user to group, then change permission level" do
+ set_user_from_auth :admin
+ grp = Group.create!(owner_uuid: system_user_uuid, group_class: "role")
+ col = Collection.create!(owner_uuid: grp.uuid)
+ assert_empty Collection.readable_by(users(:active)).where(uuid: col.uuid)
+ assert_empty User.readable_by(users(:active)).where(uuid: users(:project_viewer).uuid)
+
+ l1 = Link.create!(tail_uuid: users(:active).uuid,
+ head_uuid: grp.uuid,
+ link_class: 'permission',
+ name: 'can_manage')
+ l2 = Link.create!(tail_uuid: grp.uuid,
+ head_uuid: users(:active).uuid,
+ link_class: 'permission',
+ name: 'can_read')
+
+ assert Collection.readable_by(users(:active)).where(uuid: col.uuid).first
+ assert users(:active).can?(read: col.uuid)
+ assert users(:active).can?(write: col.uuid)
+ assert users(:active).can?(manage: col.uuid)
+
+ l1.name = 'can_read'
+ l1.save!
+
+ assert Collection.readable_by(users(:active)).where(uuid: col.uuid).first
+ assert users(:active).can?(read: col.uuid)
+ assert !users(:active).can?(write: col.uuid)
+ assert !users(:active).can?(manage: col.uuid)
+
+ l1.name = 'can_write'
+ l1.save!
+
+ assert Collection.readable_by(users(:active)).where(uuid: col.uuid).first
+ assert users(:active).can?(read: col.uuid)
+ assert users(:active).can?(write: col.uuid)
+ assert !users(:active).can?(manage: col.uuid)
+ end
+
+
+ test "add user to group, then add overlapping permission link to group" do
+ set_user_from_auth :admin
+ grp = Group.create!(owner_uuid: system_user_uuid, group_class: "role")
+ col = Collection.create!(owner_uuid: grp.uuid)
+ assert_empty Collection.readable_by(users(:active)).where(uuid: col.uuid)
+ assert_empty User.readable_by(users(:active)).where(uuid: users(:project_viewer).uuid)
+
+ l1 = Link.create!(tail_uuid: users(:active).uuid,
+ head_uuid: grp.uuid,
+ link_class: 'permission',
+ name: 'can_manage')
+ l2 = Link.create!(tail_uuid: grp.uuid,
+ head_uuid: users(:active).uuid,
+ link_class: 'permission',
+ name: 'can_read')
+
+ assert Collection.readable_by(users(:active)).where(uuid: col.uuid).first
+ assert users(:active).can?(read: col.uuid)
+ assert users(:active).can?(write: col.uuid)
+ assert users(:active).can?(manage: col.uuid)
+
+ l3 = Link.create!(tail_uuid: users(:active).uuid,
+ head_uuid: grp.uuid,
+ link_class: 'permission',
+ name: 'can_read')
+
+ assert Collection.readable_by(users(:active)).where(uuid: col.uuid).first
+ assert users(:active).can?(read: col.uuid)
+ assert users(:active).can?(write: col.uuid)
+ assert users(:active).can?(manage: col.uuid)
+
+ l3.destroy!
+
+ assert Collection.readable_by(users(:active)).where(uuid: col.uuid).first
+ assert users(:active).can?(read: col.uuid)
+ assert users(:active).can?(write: col.uuid)
+ assert users(:active).can?(manage: col.uuid)
+ end
+
+
+ test "add user to group, then add overlapping permission link to subproject" do
+ set_user_from_auth :admin
+ grp = Group.create!(owner_uuid: system_user_uuid, group_class: "project")
+ prj = Group.create!(owner_uuid: grp.uuid, group_class: "project")
+ assert_empty Group.readable_by(users(:active)).where(uuid: prj.uuid)
+ assert_empty User.readable_by(users(:active)).where(uuid: users(:project_viewer).uuid)
+
+ l1 = Link.create!(tail_uuid: users(:active).uuid,
+ head_uuid: grp.uuid,
+ link_class: 'permission',
+ name: 'can_manage')
+ l2 = Link.create!(tail_uuid: grp.uuid,
+ head_uuid: users(:active).uuid,
+ link_class: 'permission',
+ name: 'can_read')
+
+ assert Group.readable_by(users(:active)).where(uuid: prj.uuid).first
+ assert users(:active).can?(read: prj.uuid)
+ assert users(:active).can?(write: prj.uuid)
+ assert users(:active).can?(manage: prj.uuid)
+
+ l3 = Link.create!(tail_uuid: grp.uuid,
+ head_uuid: prj.uuid,
+ link_class: 'permission',
+ name: 'can_read')
+
+ assert Group.readable_by(users(:active)).where(uuid: prj.uuid).first
+ assert users(:active).can?(read: prj.uuid)
+ assert users(:active).can?(write: prj.uuid)
+ assert users(:active).can?(manage: prj.uuid)
+
+ l3.destroy!
+
+ assert Group.readable_by(users(:active)).where(uuid: prj.uuid).first
+ assert users(:active).can?(read: prj.uuid)
+ assert users(:active).can?(write: prj.uuid)
+ assert users(:active).can?(manage: prj.uuid)
+ end
end
if auto_admin_first_user_config
# This test requires no admin users exist (except for the system user)
- users(:admin).delete
+ act_as_system_user do
+ users(:admin).update_attributes!(is_admin: false)
+ end
@all_users = User.where("uuid not like '%-000000000000000'").where(:is_admin => true)
assert_equal 0, @all_users.count, "No admin users should exist (except for the system user)"
end
vm = VirtualMachine.create
- # Set up the bogus Link
- bad_uuid = 'zzzzz-tpzed-xyzxyzxyzxyzxyz'
-
- resp_link = Link.create ({tail_uuid: email, link_class: 'permission',
- name: 'can_login', head_uuid: bad_uuid})
- resp_link.save(validate: false)
-
- verify_link resp_link, 'permission', 'can_login', email, bad_uuid
-
response = user.setup(repo_name: 'foo/testrepo',
vm_uuid: vm.uuid)
err := filepath.Walk(v.Root, func(path string, info os.FileInfo, err error) error {
if err != nil {
v.logger.WithError(err).Errorf("EmptyTrash: filepath.Walk(%q) failed", path)
+ // Don't give up -- keep walking other
+ // files/dirs
return nil
+ } else if !info.Mode().IsDir() {
+ todo <- dirent{path, info}
+ return nil
+ } else if path == v.Root || blockDirRe.MatchString(info.Name()) {
+ // Descend into a directory that we might have
+ // put trash in.
+ return nil
+ } else {
+ // Don't descend into other dirs.
+ return filepath.SkipDir
}
- todo <- dirent{path, info}
- return nil
})
close(todo)
wg.Wait()
c.Check(err, check.IsNil)
c.Check(stats(), check.Matches, `.*"FlockOps":2,.*`)
}
+
+func (s *UnixVolumeSuite) TestSkipUnusedDirs(c *check.C) {
+ vol := s.newTestableUnixVolume(c, s.cluster, arvados.Volume{Replication: 1}, s.metrics, false)
+
+ err := os.Mkdir(vol.UnixVolume.Root+"/aaa", 0777)
+ c.Assert(err, check.IsNil)
+ err = os.Mkdir(vol.UnixVolume.Root+"/.aaa", 0777) // EmptyTrash should not look here
+ c.Assert(err, check.IsNil)
+ deleteme := vol.UnixVolume.Root + "/aaa/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.trash.1"
+ err = ioutil.WriteFile(deleteme, []byte{1, 2, 3}, 0777)
+ c.Assert(err, check.IsNil)
+ skipme := vol.UnixVolume.Root + "/.aaa/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa.trash.1"
+ err = ioutil.WriteFile(skipme, []byte{1, 2, 3}, 0777)
+ c.Assert(err, check.IsNil)
+ vol.EmptyTrash()
+
+ _, err = os.Stat(skipme)
+ c.Check(err, check.IsNil)
+
+ _, err = os.Stat(deleteme)
+ c.Check(err, check.NotNil)
+ c.Check(os.IsNotExist(err), check.Equals, true)
+}