Encryption: Cache new DEKs (only) after a caution period (#60664)

* Encryption: Cache new DEKs (only) after commit

* Fix typo

* Update secrets manager tests with new failing case

* Update secrets manager tests with new clarifications (comments)

* Correct broken method calls

* Unify methods

* Cache data keys only after a caution period

* Caution period for data keys caching only for encrypt ops
This commit is contained in:
Joan López de la Franca Beltran
2023-01-26 10:54:31 +01:00
committed by GitHub
parent 3a442610d2
commit c4e067d49d
7 changed files with 301 additions and 114 deletions
+170 -13
View File
@@ -2,6 +2,7 @@ package manager
import (
"context"
"errors"
"testing"
"time"
@@ -17,12 +18,14 @@ import (
"github.com/grafana/grafana/pkg/services/kmsproviders/osskmsproviders"
"github.com/grafana/grafana/pkg/services/secrets"
"github.com/grafana/grafana/pkg/services/secrets/database"
"github.com/grafana/grafana/pkg/services/sqlstore"
"github.com/grafana/grafana/pkg/setting"
"github.com/grafana/grafana/pkg/util"
)
func TestSecretsService_EnvelopeEncryption(t *testing.T) {
store := database.ProvideSecretsStore(db.InitTestDB(t))
testDB := db.InitTestDB(t)
store := database.ProvideSecretsStore(testDB)
svc := SetupTestService(t, store)
ctx := context.Background()
@@ -82,7 +85,8 @@ func TestSecretsService_EnvelopeEncryption(t *testing.T) {
}
func TestSecretsService_DataKeys(t *testing.T) {
store := database.ProvideSecretsStore(db.InitTestDB(t))
testDB := db.InitTestDB(t)
store := database.ProvideSecretsStore(testDB)
ctx := context.Background()
dataKey := &secrets.DataKey{
@@ -160,7 +164,8 @@ func TestSecretsService_DataKeys(t *testing.T) {
func TestSecretsService_UseCurrentProvider(t *testing.T) {
t.Run("When encryption_provider is not specified explicitly, should use 'secretKey' as a current provider", func(t *testing.T) {
svc := SetupTestService(t, database.ProvideSecretsStore(db.InitTestDB(t)))
testDB := db.InitTestDB(t)
svc := SetupTestService(t, database.ProvideSecretsStore(testDB))
assert.Equal(t, secrets.ProviderID("secretKey.v1"), svc.currentProviderID)
})
@@ -187,7 +192,8 @@ func TestSecretsService_UseCurrentProvider(t *testing.T) {
features := featuremgmt.WithFeatures()
kms := newFakeKMS(osskmsproviders.ProvideService(encryptionService, settings, features))
secretStore := database.ProvideSecretsStore(db.InitTestDB(t))
testDB := db.InitTestDB(t)
secretStore := database.ProvideSecretsStore(testDB)
secretsService, err := ProvideSecretsService(
secretStore,
@@ -261,8 +267,8 @@ func (f *fakeKMS) Provide() (map[secrets.ProviderID]secrets.Provider, error) {
func TestSecretsService_Run(t *testing.T) {
ctx := context.Background()
sql := db.InitTestDB(t)
store := database.ProvideSecretsStore(sql)
testDB := db.InitTestDB(t)
store := database.ProvideSecretsStore(testDB)
svc := SetupTestService(t, store)
t.Run("should stop with no error once the context's finished", func(t *testing.T) {
@@ -274,16 +280,26 @@ func TestSecretsService_Run(t *testing.T) {
})
t.Run("should trigger cache clean up", func(t *testing.T) {
// Encrypt to ensure there's a data encryption key generated
_, err := svc.Encrypt(ctx, []byte("grafana"), secrets.WithoutScope())
restoreTimeNowAfterTestExec(t)
// Encrypt to force data encryption key generation
encrypted, err := svc.Encrypt(ctx, []byte("grafana"), secrets.WithoutScope())
require.NoError(t, err)
// Ten minutes later (after caution period)
// Look SecretsService.cacheDataKey for more details.
now = func() time.Time { return time.Now().Add(10 * time.Minute) }
// Decrypt to ensure data encryption key is cached
_, err = svc.Decrypt(ctx, encrypted)
require.NoError(t, err)
// Data encryption key cache should contain one element
require.Len(t, svc.dataKeyCache.byId, 1)
require.Len(t, svc.dataKeyCache.byLabel, 1)
t.Cleanup(func() { now = time.Now })
now = func() time.Time { return time.Now().Add(10 * time.Minute) }
// Twenty minutes later (after caution period + cache ttl)
now = func() time.Time { return time.Now().Add(20 * time.Minute) }
ctx, cancel := context.WithTimeout(ctx, 1*time.Second)
defer cancel()
@@ -301,8 +317,8 @@ func TestSecretsService_Run(t *testing.T) {
func TestSecretsService_ReEncryptDataKeys(t *testing.T) {
ctx := context.Background()
sql := db.InitTestDB(t)
store := database.ProvideSecretsStore(sql)
testDB := db.InitTestDB(t)
store := database.ProvideSecretsStore(testDB)
svc := SetupTestService(t, store)
// Encrypt to generate data encryption key
@@ -326,6 +342,12 @@ func TestSecretsService_ReEncryptDataKeys(t *testing.T) {
})
t.Run("data keys cache should be invalidated", func(t *testing.T) {
restoreTimeNowAfterTestExec(t)
// Ten minutes later (after caution period)
// Look SecretsService.cacheDataKey for more details.
now = func() time.Time { return time.Now().Add(10 * time.Minute) }
// Decrypt to ensure data key is cached
_, err := svc.Decrypt(ctx, ciphertext)
require.NoError(t, err)
@@ -342,7 +364,8 @@ func TestSecretsService_ReEncryptDataKeys(t *testing.T) {
func TestSecretsService_Decrypt(t *testing.T) {
ctx := context.Background()
store := database.ProvideSecretsStore(db.InitTestDB(t))
testDB := db.InitTestDB(t)
store := database.ProvideSecretsStore(testDB)
t.Run("empty payload should fail", func(t *testing.T) {
svc := SetupTestService(t, store)
@@ -401,3 +424,137 @@ func TestSecretsService_Decrypt(t *testing.T) {
assert.Equal(t, []byte("grafana"), decrypted)
})
}
func TestIntegration_SecretsService(t *testing.T) {
if testing.Short() {
t.Skip("skipping integration test")
}
ctx := context.Background()
someData := []byte(`some-data`)
tcs := map[string]func(*testing.T, *sqlstore.SQLStore, *SecretsService){
"regular": func(t *testing.T, _ *sqlstore.SQLStore, svc *SecretsService) {
// We encrypt some data normally, no transactions implied.
_, err := svc.Encrypt(ctx, someData, secrets.WithoutScope())
require.NoError(t, err)
},
"within successful InTransaction": func(t *testing.T, store *sqlstore.SQLStore, svc *SecretsService) {
require.NoError(t, store.InTransaction(ctx, func(ctx context.Context) error {
// We encrypt some data within a transaction that shares the db session.
_, err := svc.Encrypt(ctx, someData, secrets.WithoutScope())
require.NoError(t, err)
// And the transition succeeds.
return nil
}))
},
"within unsuccessful InTransaction": func(t *testing.T, store *sqlstore.SQLStore, svc *SecretsService) {
require.NotNil(t, store.InTransaction(ctx, func(ctx context.Context) error {
// We encrypt some data within a transaction that shares the db session.
_, err := svc.Encrypt(ctx, someData, secrets.WithoutScope())
require.NoError(t, err)
// But the transaction fails.
return errors.New("error")
}))
},
"within unsuccessful InTransaction (plus forced db fetch)": func(t *testing.T, store *sqlstore.SQLStore, svc *SecretsService) {
require.NotNil(t, store.InTransaction(ctx, func(ctx context.Context) error {
// We encrypt some data within a transaction that shares the db session.
encrypted, err := svc.Encrypt(ctx, someData, secrets.WithoutScope())
require.NoError(t, err)
// At this point the data key is not cached yet because
// the transaction haven't been committed yet,
// and won't, so we do a decrypt operation within the
// transaction to force the data key to be
// (potentially) cached (it shouldn't to prevent issues).
decrypted, err := svc.Decrypt(ctx, encrypted)
require.NoError(t, err)
assert.Equal(t, someData, decrypted)
// But the transaction fails.
return errors.New("error")
}))
},
"within successful WithTransactionalDbSession": func(t *testing.T, store *sqlstore.SQLStore, svc *SecretsService) {
require.NoError(t, store.WithTransactionalDbSession(ctx, func(sess *sqlstore.DBSession) error {
// We encrypt some data within a transaction that does not share the db session.
_, err := svc.Encrypt(ctx, someData, secrets.WithoutScope())
require.NoError(t, err)
// And the transition succeeds.
return nil
}))
},
"within unsuccessful WithTransactionalDbSession": func(t *testing.T, store *sqlstore.SQLStore, svc *SecretsService) {
require.NotNil(t, store.WithTransactionalDbSession(ctx, func(sess *sqlstore.DBSession) error {
// We encrypt some data within a transaction that does not share the db session.
_, err := svc.Encrypt(ctx, someData, secrets.WithoutScope())
require.NoError(t, err)
// But the transaction fails.
return errors.New("error")
}))
},
"within unsuccessful WithTransactionalDbSession (plus forced db fetch)": func(t *testing.T, store *sqlstore.SQLStore, svc *SecretsService) {
require.NotNil(t, store.WithTransactionalDbSession(ctx, func(sess *sqlstore.DBSession) error {
// We encrypt some data within a transaction that does not share the db session.
encrypted, err := svc.Encrypt(ctx, someData, secrets.WithoutScope())
require.NoError(t, err)
// At this point the data key is not cached yet because
// the transaction haven't been committed yet,
// and won't, so we do a decrypt operation within the
// transaction to force the data key to be
// (potentially) cached (it shouldn't to prevent issues).
decrypted, err := svc.Decrypt(ctx, encrypted)
require.NoError(t, err)
assert.Equal(t, someData, decrypted)
// But the transaction fails.
return errors.New("error")
}))
},
}
for name, tc := range tcs {
t.Run(name, func(t *testing.T) {
testDB := db.InitTestDB(t)
svc := SetupTestService(t, database.ProvideSecretsStore(testDB))
// Here's what actually matters and varies on each test: look at the test case name.
//
// For historical reasons, and in an old implementation, when a successful encryption
// operation happened within an unsuccessful transaction, the data key was used to be
// cached in memory for the next encryption operations, which caused some data to be
// encrypted with a data key that haven't actually been persisted into the database.
tc(t, testDB, svc)
// Therefore, the data encrypted after this point, become unrecoverable after a restart.
// So, the different test cases here are there to prevent that from happening again
// in the future, whatever it is what happens.
// So, we proceed with an encryption operation:
toEncrypt := []byte(`data-to-encrypt`)
encrypted, err := svc.Encrypt(ctx, toEncrypt, secrets.WithoutScope())
require.NoError(t, err)
// We simulate an instance restart. So, there's no data in the in-memory cache.
svc.dataKeyCache.flush()
// And then, we MUST still be able to decrypt the previously encrypted data:
decrypted, err := svc.Decrypt(ctx, encrypted)
require.NoError(t, err)
assert.Equal(t, toEncrypt, decrypted)
})
}
}
// Use this function at the beginning of those tests
// that manipulates 'now', so it'll leave it in a
// correct state once test execution finishes.
func restoreTimeNowAfterTestExec(t *testing.T) {
t.Helper()
t.Cleanup(func() { now = time.Now })
}