@@ -283,8 +283,22 @@ static struct crypto_alg *crypto_larval_lookup(const char *name, u32 type,
if (!IS_ERR_OR_NULL(alg) && crypto_is_larval(alg))
alg = crypto_larval_wait(alg);
- else if (!alg)
- alg = crypto_larval_add(name, type, mask);
+ else if (!alg) {
+ /*
+ * Only add a lookup larval if the request is for a
+ * tested algorithm, everything else makes no sense.
+ */
+ bool tested = type & mask & CRYPTO_ALG_TESTED;
+
+ if (!(mask & CRYPTO_ALG_TESTED)) {
+ WARN_ON_ONCE((type & CRYPTO_ALG_TESTED));
+ tested = true;
+ }
+ if (tested)
+ alg = crypto_larval_add(name, type, mask);
+ else
+ alg = ERR_PTR(-ENOENT);
+ }
return alg;
}
As of now, the only valid (and existent) usecase for requesting an algorithm with !(type & CRYPTO_ALG_TESTED) and (mask & CRYPTO_ALG_TESTED) via crypto_alg_mod_lookup() is the testmgr invoked upon a CRYPTO_MSG_ALG_REGISTER notification. However, in this case it is expected and required that the lookup returns the actual crypto_alg instance subject to testing. If OTOH we were to assume for a moment that the crypto_larval_lookup() called from crypto_alg_mod_lookup() would allocate and return a lookup larval (which cannot happen, except for perhaps in some corner case like with concurrent crypto_alg unregistrations), the subsequent wait in crypto_alg_mod_lookup() would timeout anyway. Make crypto_larval_wait() skip the lookup larval allocation if the specified values of mask and type indicate a lookup request for some crypto_alg instance still not in CRYPTO_ALG_TESTED state and the search for such one failed. Instead, fail the call with -ENOENT in this case. This reduces the number of possible ->mask and ->type configurations with respect to CRYPTO_ALG_TESTED for pending lookup larvals and will enable further cleanup with subsequent patches. There is no change in functionality. Signed-off-by: Nicolai Stange <nstange@suse.de> --- crypto/api.c | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-)