Message-ID: <ZpUrPJ0hxqc7LaAt@tissot.1015granger.net>
Date: Mon, 15 Jul 2024 09:59:24 -0400
From: Chuck Lever <chuck.lever@...cle.com>
To: cuigaosheng <cuigaosheng1@...wei.com>
Cc: Neil Brown <neilb@...e.de>, Trond Myklebust <trondmy@...nel.org>,
        Anna Schumaker <anna@...nel.org>, Jeff Layton <jlayton@...nel.org>,
        Olga Kornievskaia <kolga@...app.com>, Dai Ngo <dai.ngo@...cle.com>,
        Tom Talpey <tom@...pey.com>, "David S. Miller" <davem@...emloft.net>,
        Eric Dumazet <edumazet@...gle.com>, Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>, Simon Horman <horms@...nel.org>,
        Linux NFS Mailing List <linux-nfs@...r.kernel.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: [PATCH -next,v2] gss_krb5: refactor code to return correct
 PTR_ERR in krb5_DK

On Mon, Jul 15, 2024 at 09:44:30AM +0800, cuigaosheng wrote:
> But crypto_sync_skcipher_setkey maybe return -ENOMEM, Should this be
> modified?

Auditing this path is a bit challenging. The obvious memory failure
case is skcipher_setkey_unaligned(), but that is called only if
the cipher does not provide its own ->setkey method. (Did you see a
failure case that I missed?)

So you'll have to generate a list of ciphers that krb5_DK() uses
(which is only a few) and then check the ->setkey method of each
of those ciphers.

My guess is the skcipher_setkey fallback isn't used for any of the
ciphers that SunRPC GSS currently cares about.


> On 2024/7/15 0:18, Chuck Lever III wrote:
> > 
> > > On Jul 14, 2024, at 12:31 AM, NeilBrown <neilb@...e.de> wrote:
> > > 
> > > On Sat, 13 Jul 2024, Chuck Lever wrote:
> > > > On Fri, Jul 12, 2024 at 09:39:08AM -0400, Chuck Lever wrote:
> > > > > On Fri, Jul 12, 2024 at 03:24:23PM +0800, Gaosheng Cui wrote:
> > > > > > Refactor the code in krb5_DK to return PTR_ERR when an error occurs.
> > > > > My understanding of the current code is that if either
> > > > > crypto_alloc_sync_skcipher() or crypto_sync_skcipher_blocksize()
> > > > > fails, then krb5_DK() returns -EINVAL. At the only call site for
> > > > > krb5_DK(), that return code is unconditionally discarded. Thus I
> > > > > don't see that the proposed change is necessary or improves
> > > > > anything.
> > > > My understanding is wrong  ;-)
> > > True, but I think your conclusion was correct.
> > > 
> > > krb5_DK() returns zero or -EINVAL.
> > > It is only used by krb5_derive_key_v2(), which returns zero or -EINVAL,
> > > or -ENOMEM.
> > These are really the only three interesting return codes.
> > Leaking other error codes to callers is not desirable, IMO.
> > 
> > But looking at the current implementation of
> > crypto_alloc_sync_skcipher(), it returns either
> > ERR_PTR(-EINVAL) or a valid pointer; it doesn't return any
> > other error value. Since it never returns -ENOMEM, there
> > still doesn't seem to be a technical reason for modifying
> > krb5_DK() to pass errors through.
> > 
> > 
> > > krb4_derive_key_v2() is only used as a ->derive_key() method.
> > > This is called from krb5_derive_key(), and various unit tests in
> > > gss_krb5_tests.c
> > > 
> > > krb5_derive_key() is only called in gss_krb5_mech.c, and each call site
> > > is of the form:
> > >   if (krb5_derive_key(...)) goto out;
> > > so it doesn't matter what error is returned.
> > > 
> > > The unit test calls are all followed by
> > > KUNIT_ASSERT_EQ(test, err, 0);
> > > so the only place the err is used is (presumably) in failure reports
> > > from the unit tests.
> > > 
> > > So the proposed change seems unnecessary from a practical perspective.
> > > 
> > > Maybe it is justified from an aesthetic perspective, but I think that
> > > should be clearly stated in the commit message.  e.g.
> > > 
> > >   This change has no practical effect as all non-zero error statuses
> > >   are treated equally, however the distinction between EINVAL and ENOMEM
> > >   may be relevant at some future time and it seems cleaner to maintain
> > >   the distinction.
> > > 
> > > NeilBrown
> > > 
> > > 
> > > > The return code isn't discarded. A non-zero return code from
> > > > krb5_DK() is carried back up the call stack. The logic in
> > > > krb5_derive_key_v2() does not use the kernel's usual error flow
> > > > form, so I missed this.
> > > > 
> > > > However, it still isn't clear to me why the error behavior here
> > > > needs to change. It's possible, for example, that -EINVAL is
> > > > perfectly adequate to indicate when sync_skcipher() can't find the
> > > > specified encryption algorithm (gk5e->encrypt_name).
> > > > 
> > > > Specifying the wrong encryption type: -EINVAL. That makes sense.
> > > > 
> > > > 
> > > > > > Signed-off-by: Gaosheng Cui <cuigaosheng1@...wei.com>
> > > > > > ---
> > > > > > v2: Update IS_ERR to PTR_ERR, thanks very much!
> > > > > > net/sunrpc/auth_gss/gss_krb5_keys.c | 8 ++++++--
> > > > > > 1 file changed, 6 insertions(+), 2 deletions(-)
> > > > > > 
> > > > > > diff --git a/net/sunrpc/auth_gss/gss_krb5_keys.c b/net/sunrpc/auth_gss/gss_krb5_keys.c
> > > > > > index 4eb19c3a54c7..5ac8d06ab2c0 100644
> > > > > > --- a/net/sunrpc/auth_gss/gss_krb5_keys.c
> > > > > > +++ b/net/sunrpc/auth_gss/gss_krb5_keys.c
> > > > > > @@ -164,10 +164,14 @@ static int krb5_DK(const struct gss_krb5_enctype *gk5e,
> > > > > > goto err_return;
> > > > > > 
> > > > > > cipher = crypto_alloc_sync_skcipher(gk5e->encrypt_name, 0, 0);
> > > > > > - if (IS_ERR(cipher))
> > > > > > + if (IS_ERR(cipher)) {
> > > > > > + ret = PTR_ERR(cipher);
> > > > > > goto err_return;
> > > > > > + }
> > > > > > +
> > > > > > blocksize = crypto_sync_skcipher_blocksize(cipher);
> > > > > > - if (crypto_sync_skcipher_setkey(cipher, inkey->data, inkey->len))
> > > > > > + ret = crypto_sync_skcipher_setkey(cipher, inkey->data, inkey->len);
> > > > > > + if (ret)
> > > > > > goto err_free_cipher;
> > > > > > 
> > > > > > ret = -ENOMEM;
> > > > > > -- 
> > > > > > 2.25.1
> > > > > > 
> > > > > -- 
> > > > > Chuck Lever
> > > > > 
> > > > -- 
> > > > Chuck Lever
> > > > 
> > --
> > Chuck Lever
> > 
> > 

-- 
Chuck Lever

Powered by blists - more mailing lists