On 3/11/2024 7:57 PM, Eric Biggers wrote:
On Wed, Mar 06, 2024 at 03:34:40PM -0800, Fan Wu wrote:
fsverity represents a mechanism to support both integrity and
authenticity protection of a file, supporting both signed and unsigned
digests.
An LSM which controls access to a resource based on authenticity and
integrity of said resource, can then use this data to make an informed
decision on the authorization (provided by the LSM's policy) of said
claim.
This effectively allows the extension of a policy enforcement layer in
LSM for fsverity, allowing for more granular control of how a
particular authenticity claim can be used. For example, "all (built-in)
signed fsverity files should be allowed to execute, but only these
hashes are allowed to be loaded as kernel modules".
This enforcement must be done in kernel space, as a userspace only
solution would fail a simple litmus test: Download a self-contained
malicious binary that never touches the userspace stack. This
binary would still be able to execute.
Signed-off-by: Deven Bowers <deven.desai@xxxxxxxxxxxxxxxxxxx>
Signed-off-by: Fan Wu <wufan@xxxxxxxxxxxxxxxxxxx>
As I've said before, this commit message needs some work. It currently doesn't
say anything about what the patch actually does.
Apologies, I must have missed your previous comment. I appreciate your
feedback and will ensure that the commit message clearly describes what
the patch does in the next version. Thank you for pointing this out.
BTW, please make sure you're Cc'ing the fsverity mailing list
(fsverity@xxxxxxxxxxxxxxx), not fscrypt (linux-fscrypt@xxxxxxxxxxxxxxx).
Thanks for the info.
diff --git a/Documentation/filesystems/fsverity.rst b/Documentation/filesystems/fsverity.rst
index 13e4b18e5dbb..64618a6141ab 100644
--- a/Documentation/filesystems/fsverity.rst
+++ b/Documentation/filesystems/fsverity.rst
@@ -461,7 +461,9 @@ Enabling this option adds the following:
3. A new sysctl "fs.verity.require_signatures" is made available.
When set to 1, the kernel requires that all verity files have a
- correctly signed digest as described in (2).
+ correctly signed digest as described in (2). Note that verification
+ happens as long as the file's signature exists regardless the state of
+ "fs.verity.require_signatures".
The data that the signature as described in (2) must be a signature of
is the fs-verity file digest in the following format::
Doesn't anything else in this file need to be updated to document the IPE
support?
Yes, I can add more details of IPE support in the built-in signature
section.
diff --git a/fs/verity/open.c b/fs/verity/open.c
index 6c31a871b84b..f917023255c8 100644
--- a/fs/verity/open.c
+++ b/fs/verity/open.c
@@ -8,6 +8,7 @@
#include "fsverity_private.h"
#include <linux/mm.h>
+#include <linux/security.h>
#include <linux/slab.h>
static struct kmem_cache *fsverity_info_cachep;
@@ -172,12 +173,28 @@ static int compute_file_digest(const struct fsverity_hash_alg *hash_alg,
return err;
}
+#ifdef CONFIG_FS_VERITY_BUILTIN_SIGNATURES
+static int fsverity_inode_setsecurity(struct inode *inode,
+ const struct fsverity_descriptor *desc)
+{
+ return security_inode_setsecurity(inode, FS_VERITY_INODE_SEC_NAME,
+ desc->signature,
+ le32_to_cpu(desc->sig_size), 0);
Please call it something like FS_VERITY_INODE_BUILTIN_SIG to make it clear that
it's the builtin signature.
Thanks for the suggestion, I will take the name.
+}
+#else
+static inline int fsverity_inode_setsecurity(struct inode *inode,
+ const struct fsverity_descriptor *desc)
+{
+ return 0;
+}
+#endif /* CONFIG_IPE_PROP_FS_VERITY*/
The above comment mentions CONFIG_IPE_PROP_FS_VERITY, but it doesn't appear
anywhere else in the patch.
The CONFIG_IPE_PROP_FS_VERITY KCONFIG switch is introduced in a
subsequent patch. Merging these two patches might not be the best
approach. I can switch to using CONFIG_FS_VERITY_BUILTIN_SIGNATURES
instead, which may be more appropriate for the current context.
+struct fsverity_info *fsverity_create_info(struct inode *inode,
struct fsverity_descriptor *desc)
{
struct fsverity_info *vi;
@@ -242,6 +259,13 @@ struct fsverity_info *fsverity_create_info(const struct inode *inode,
spin_lock_init(&vi->hash_page_init_lock);
}
+ err = fsverity_inode_setsecurity(inode, desc);
+ if (err == -EOPNOTSUPP)
+ err = 0;
What is the "err == -EOPNOTSUPP" case intended to handle?
The -EOPNOTSUPP is designed to signal situations where the called LSM
hook does not support associating the passed name with a value, but the
hook call itself does not encounter any errors. Also, -EOPNOTSUPP is the
default return value of security_inode_setsecurity when no LSM has
registered the hook. In summary, it indicates that no hook call
succeeded, but no critical error occurred.
diff --git a/fs/verity/signature.c b/fs/verity/signature.c
index 90c07573dd77..42f58f4e45d0 100644
--- a/fs/verity/signature.c
+++ b/fs/verity/signature.c
@@ -41,7 +41,9 @@ static struct key *fsverity_keyring;
* @sig_size: size of signature in bytes, or 0 if no signature
*
* If the file includes a signature of its fs-verity file digest, verify it
- * against the certificates in the fs-verity keyring.
+ * against the certificates in the fs-verity keyring. Note that verification
+ * happens as long as the file's signature exists regardless the state of
+ * fsverity_require_signatures.
Can you please make this mention explicitly that the LSM hook is relying on that
behavior?
Sure, I can add this info.
-Fan
- Eric
