On Wed, Nov 22, 2017 at 08:55:14AM +0000, Ard Biesheuvel wrote: > Hello Corentin, > > On 22 November 2017 at 08:08, Corentin Labbe <clabbe@xxxxxxxxxxxx> wrote: > > When a cipher fail > > fails > > > to register in aes_init(), the error path go thought > > goes through > > > aes_exit() then crypto_unregister_skciphers(). > > Since aes_exit calls also crypto_unregister_skcipher, this trigger a > > triggers > > > refcount_t: underflow; use-after-free. > > > > Signed-off-by: Corentin Labbe <clabbe@xxxxxxxxxxxx> > > --- > > arch/arm64/crypto/aes-glue.c | 5 ++++- > > 1 file changed, 4 insertions(+), 1 deletion(-) > > > > diff --git a/arch/arm64/crypto/aes-glue.c b/arch/arm64/crypto/aes-glue.c > > index 998ba519a026..9e42ec96243e 100644 > > --- a/arch/arm64/crypto/aes-glue.c > > +++ b/arch/arm64/crypto/aes-glue.c > > @@ -664,7 +664,10 @@ static int __init aes_init(void) > > return 0; > > > > unregister_simds: > > - aes_exit(); > > + for (i = 0; i < ARRAY_SIZE(aes_simd_algs); i++) > > + if (aes_simd_algs[i]) > > + simd_skcipher_free(aes_simd_algs[i]); > > + crypto_unregister_shashes(mac_algs, ARRAY_SIZE(mac_algs)); > > unregister_ciphers: > > crypto_unregister_skciphers(aes_algs, ARRAY_SIZE(aes_algs)); > > return err; > > -- > > 2.13.6 > > > > > > > Would this also fix it? > > diff --git a/arch/arm64/crypto/aes-glue.c b/arch/arm64/crypto/aes-glue.c > index 998ba519a026..2fa850e86aa8 100644 > --- a/arch/arm64/crypto/aes-glue.c > +++ b/arch/arm64/crypto/aes-glue.c > @@ -665,6 +665,7 @@ static int __init aes_init(void) > > unregister_simds: > aes_exit(); > + return err; > unregister_ciphers: > crypto_unregister_skciphers(aes_algs, ARRAY_SIZE(aes_algs)); > return err; Yes it is better. I will send a v2 today. Regards