[PATCH] fsverity: don't check builtin signatures when require_signatures=0

[PATCH] fsverity: don't check builtin signatures when require_signatures=0

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

An issue that arises when migrating from builtin signatures to userspace
signatures is that existing files that have builtin signatures cannot be
opened unless either CONFIG_FS_VERITY_BUILTIN_SIGNATURES is disabled or
the signing certificate is left in the .fs-verity keyring.

Since builtin signatures provide no security benefit when
fs.verity.require_signatures=0 anyway, let's just skip the signature
verification in this case.

Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
Cc: <stable@vger.kernel.org> # v5.4+
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 fs/verity/signature.c | 18 ++++++++++++++++--
 1 file changed, 16 insertions(+), 2 deletions(-)

diff --git a/fs/verity/signature.c b/fs/verity/signature.c
index 143a530a80088..dc6935701abda 100644
--- a/fs/verity/signature.c
+++ b/fs/verity/signature.c
@@ -13,8 +13,8 @@
 #include <linux/verification.h>
 
 /*
- * /proc/sys/fs/verity/require_signatures
- * If 1, all verity files must have a valid builtin signature.
+ * /proc/sys/fs/verity/require_signatures.  If 1, then builtin signatures are
+ * verified and all verity files must have a valid builtin signature.
  */
 static int fsverity_require_signatures;
 
@@ -54,6 +54,20 @@ int fsverity_verify_signature(const struct fsverity_info *vi,
 		return 0;
 	}
 
+	/*
+	 * If require_signatures=0, don't verify builtin signatures.
+	 * Originally, builtin signatures were verified opportunistically in
+	 * this case.  However, no security property is possible when
+	 * require_signatures=0 anyway.  Skipping the builtin signature
+	 * verification makes it easier to migrate existing files from builtin
+	 * signature verification to userspace signature verification.
+	 */
+	if (!fsverity_require_signatures) {
+		fsverity_warn(inode,
+			      "Not checking builtin signature due to require_signatures=0");
+		return 0;
+	}
+
 	d = kzalloc(sizeof(*d) + hash_alg->digest_size, GFP_KERNEL);
 	if (!d)
 		return -ENOMEM;

base-commit: 479174d402bcf60789106eedc4def3957c060bad
-- 
2.38.1

Re: [PATCH] fsverity: don't check builtin signatures when require_signatures=0

[Luca Boccassi]

On Thu, 8 Dec 2022 at 03:35, Eric Biggers <ebiggers@kernel.org> wrote:
>
> From: Eric Biggers <ebiggers@google.com>
>
> An issue that arises when migrating from builtin signatures to userspace
> signatures is that existing files that have builtin signatures cannot be
> opened unless either CONFIG_FS_VERITY_BUILTIN_SIGNATURES is disabled or
> the signing certificate is left in the .fs-verity keyring.
>
> Since builtin signatures provide no security benefit when
> fs.verity.require_signatures=0 anyway, let's just skip the signature
> verification in this case.
>
> Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
> Cc: <stable@vger.kernel.org> # v5.4+
> Signed-off-by: Eric Biggers <ebiggers@google.com>
> ---
>  fs/verity/signature.c | 18 ++++++++++++++++--
>  1 file changed, 16 insertions(+), 2 deletions(-)

Acked-by: Luca Boccassi <bluca@debian.org>

Re: [PATCH] fsverity: don't check builtin signatures when require_signatures=0

[Eric Biggers]

On Thu, Dec 08, 2022 at 08:42:56PM +0000, Luca Boccassi wrote:
> On Thu, 8 Dec 2022 at 03:35, Eric Biggers <ebiggers@kernel.org> wrote:
> >
> > From: Eric Biggers <ebiggers@google.com>
> >
> > An issue that arises when migrating from builtin signatures to userspace
> > signatures is that existing files that have builtin signatures cannot be
> > opened unless either CONFIG_FS_VERITY_BUILTIN_SIGNATURES is disabled or
> > the signing certificate is left in the .fs-verity keyring.
> >
> > Since builtin signatures provide no security benefit when
> > fs.verity.require_signatures=0 anyway, let's just skip the signature
> > verification in this case.
> >
> > Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
> > Cc: <stable@vger.kernel.org> # v5.4+
> > Signed-off-by: Eric Biggers <ebiggers@google.com>
> > ---
> >  fs/verity/signature.c | 18 ++++++++++++++++--
> >  1 file changed, 16 insertions(+), 2 deletions(-)
> 
> Acked-by: Luca Boccassi <bluca@debian.org>

So if I can't apply
https://lore.kernel.org/linux-fscrypt/20221208033548.122704-1-ebiggers@kernel.org
("fsverity: mark builtin signatures as deprecated") due to IPE, wouldn't I not
be able to apply this patch either?  Surely IPE isn't depending on
fs.verity.require_signatures=1, given that it enforces the policy itself?

- Eric

Re: [PATCH] fsverity: don't check builtin signatures when require_signatures=0

[Luca Boccassi]

On Fri, 16 Dec 2022 at 21:06, Eric Biggers <ebiggers@kernel.org> wrote:
>
> On Thu, Dec 08, 2022 at 08:42:56PM +0000, Luca Boccassi wrote:
> > On Thu, 8 Dec 2022 at 03:35, Eric Biggers <ebiggers@kernel.org> wrote:
> > >
> > > From: Eric Biggers <ebiggers@google.com>
> > >
> > > An issue that arises when migrating from builtin signatures to userspace
> > > signatures is that existing files that have builtin signatures cannot be
> > > opened unless either CONFIG_FS_VERITY_BUILTIN_SIGNATURES is disabled or
> > > the signing certificate is left in the .fs-verity keyring.
> > >
> > > Since builtin signatures provide no security benefit when
> > > fs.verity.require_signatures=0 anyway, let's just skip the signature
> > > verification in this case.
> > >
> > > Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
> > > Cc: <stable@vger.kernel.org> # v5.4+
> > > Signed-off-by: Eric Biggers <ebiggers@google.com>
> > > ---
> > >  fs/verity/signature.c | 18 ++++++++++++++++--
> > >  1 file changed, 16 insertions(+), 2 deletions(-)
> >
> > Acked-by: Luca Boccassi <bluca@debian.org>
>
> So if I can't apply
> https://lore.kernel.org/linux-fscrypt/20221208033548.122704-1-ebiggers@kernel.org
> ("fsverity: mark builtin signatures as deprecated") due to IPE, wouldn't I not
> be able to apply this patch either?  Surely IPE isn't depending on
> fs.verity.require_signatures=1, given that it enforces the policy itself?

I'm not sure what you mean? Skipping verification when this syscfg is
disabled makes sense to me, as you noted it doesn't serve any purpose
in that case.

Kind regards,
Luca Boccassi

Re: [PATCH] fsverity: don't check builtin signatures when require_signatures=0

[Eric Biggers]

On Sat, Dec 17, 2022 at 02:06:04AM +0000, Luca Boccassi wrote:
> On Fri, 16 Dec 2022 at 21:06, Eric Biggers <ebiggers@kernel.org> wrote:
> >
> > On Thu, Dec 08, 2022 at 08:42:56PM +0000, Luca Boccassi wrote:
> > > On Thu, 8 Dec 2022 at 03:35, Eric Biggers <ebiggers@kernel.org> wrote:
> > > >
> > > > From: Eric Biggers <ebiggers@google.com>
> > > >
> > > > An issue that arises when migrating from builtin signatures to userspace
> > > > signatures is that existing files that have builtin signatures cannot be
> > > > opened unless either CONFIG_FS_VERITY_BUILTIN_SIGNATURES is disabled or
> > > > the signing certificate is left in the .fs-verity keyring.
> > > >
> > > > Since builtin signatures provide no security benefit when
> > > > fs.verity.require_signatures=0 anyway, let's just skip the signature
> > > > verification in this case.
> > > >
> > > > Fixes: 432434c9f8e1 ("fs-verity: support builtin file signatures")
> > > > Cc: <stable@vger.kernel.org> # v5.4+
> > > > Signed-off-by: Eric Biggers <ebiggers@google.com>
> > > > ---
> > > >  fs/verity/signature.c | 18 ++++++++++++++++--
> > > >  1 file changed, 16 insertions(+), 2 deletions(-)
> > >
> > > Acked-by: Luca Boccassi <bluca@debian.org>
> >
> > So if I can't apply
> > https://lore.kernel.org/linux-fscrypt/20221208033548.122704-1-ebiggers@kernel.org
> > ("fsverity: mark builtin signatures as deprecated") due to IPE, wouldn't I not
> > be able to apply this patch either?  Surely IPE isn't depending on
> > fs.verity.require_signatures=1, given that it enforces the policy itself?
> 
> I'm not sure what you mean? Skipping verification when this syscfg is
> disabled makes sense to me, as you noted it doesn't serve any purpose
> in that case.

Currently, fsverity builtin signatures are only useful if
fs.verity.require_signatures is set to 1 *and* userspace actually checks that
files have fsverity enabled.  However, IPE would change that if it actually gets
merged upstream, at least based on the version that was most recently sent out.
It would introduce a use of fsverity builtin signatures directly in the kernel
(https://lore.kernel.org/r/1654714889-26728-14-git-send-email-deven.desai@linux.microsoft.com
and
https://lore.kernel.org/r/1654714889-26728-15-git-send-email-deven.desai@linux.microsoft.com).

IIUC, the IPE patches add code that checks whether a file has a fsverity builtin
signature, and if so it assumes that it was verified by fs/verity/ and creates a
*boolean* file property "fsverity_signature" for IPE to operate on.

Since the IPE patches check for the presence of a builtin signature directly,
instead of indirectly by checking whether the inode has fsverity enabled at all,
there would be no need for the fs.verity.require_signatures setting with IPE.

The IPE patches do assume that the signature, if present, always gets verified
by fs/verity/.  That's what this patch would break.

Of course, for upstream I shouldn't care about breaking out-of-tree code.  So I
could apply this anyway.  But I'd at least like to be consistent.  If "fsverity:
mark builtin signatures as deprecated" isn't going to be applied because of IPE,
then I'd think this patch shouldn't be applied either, for the same reason...

- Eric