On Thu, Mar 25, 2021 at 01:13:41PM +0100, Christian Brauner wrote:

On Thu, Mar 25, 2021 at 10:53:43AM +0000, Roberto Sassu wrote:

quoted

quoted

From: Roberto Sassu
Sent: Friday, March 5, 2021 4:19 PM
With the patch to allow xattr/attr operations if a portable signature
verification fails, cp and tar can copy all xattrs/attrs so that at the
end of the process verification succeeds.

However, it might happen that the xattrs/attrs are already set to the
correct value (taken at signing time) and signature verification succeeds
before the copy has completed. For example, an archive might contains files
owned by root and the archive is extracted by root.

Then, since portable signatures are immutable, all subsequent operations
fail (e.g. fchown()), even if the operation is legitimate (does not alter
the current value).

This patch avoids this problem by reporting successful operation to user
space when that operation does not alter the current value of xattrs/attrs.

Signed-off-by: Roberto Sassu <roberto.sassu@huawei.com>
---
 security/integrity/evm/evm_main.c | 96
+++++++++++++++++++++++++++++++
 1 file changed, 96 insertions(+)

diff --git a/security/integrity/evm/evm_main.c

b/security/integrity/evm/evm_main.c
index eab536fa260f..a07516dcb920 100644

--- a/security/integrity/evm/evm_main.c
+++ b/security/integrity/evm/evm_main.c

@@ -18,6 +18,7 @@
 #include <linux/integrity.h>
 #include <linux/evm.h>
 #include <linux/magic.h>
+#include <linux/posix_acl_xattr.h>

 #include <crypto/hash.h>
 #include <crypto/hash_info.h>

@@ -328,6 +329,79 @@ static enum integrity_status

evm_verify_current_integrity(struct dentry *dentry)
 	return evm_verify_hmac(dentry, NULL, NULL, 0, NULL);
 }

+/*
+ * evm_xattr_acl_change - check if passed ACL changes the inode mode
+ * @dentry: pointer to the affected dentry
+ * @xattr_name: requested xattr
+ * @xattr_value: requested xattr value
+ * @xattr_value_len: requested xattr value length
+ *
+ * Check if passed ACL changes the inode mode, which is protected by
EVM.
+ *
+ * Returns 1 if passed ACL causes inode mode change, 0 otherwise.
+ */
+static int evm_xattr_acl_change(struct dentry *dentry, const char
*xattr_name,
+				const void *xattr_value, size_t
xattr_value_len)
+{
+	umode_t mode;
+	struct posix_acl *acl = NULL, *acl_res;
+	struct inode *inode = d_backing_inode(dentry);
+	int rc;
+
+	/* UID/GID in ACL have been already converted from user to init ns
*/
+	acl = posix_acl_from_xattr(&init_user_ns, xattr_value,
xattr_value_len);
+	if (!acl)

Based on Mimi's review, I will change this to:

if (IS_ERR_OR_NULL(acl))

quoted

+		return 1;
+
+	acl_res = acl;
+	rc = posix_acl_update_mode(&init_user_ns, inode, &mode,
&acl_res);

About this part, probably it is not correct.

I'm writing a test for this patch that checks if operations
that don't change the file mode succeed and those that
do fail.

mount-idmapped --map-mount b:3001:0:1 /mnt /mnt-idmapped
pushd /mnt
echo "test" > test-file
chown 3001 test-file
chgrp 3001 test-file
chmod 2644 test-file
<check enabled>
setfacl --set u::rw,g::r,o::r,m:r test-file (expected to succeed, caller has CAP_FSETID, so S_ISGID is not dropped)
setfacl --set u::rw,g::r,o::r,m:rw test-file (expected to fail)
pushd /mnt-idmapped
capsh --drop=cap_fsetid -- -c setfacl --set u::rw,g::r,o::r test-file (expected to succeed, caller is in the owning group of test-file, so S_ISGID is not dropped)

After adding a debug line in posix_acl_update_mode():
printk("%s: %d(%d) %d\n", __func__, in_group_p(i_gid_into_mnt(mnt_userns, inode)), __kgid_val(i_gid_into_mnt(mnt_userns, inode)), capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FSETID));

without passing mnt_userns:
[  748.262582] setfacl --set u::rw,g::r,o::r,m:r test-file
[  748.268021] posix_acl_update_mode: 0(3001) 1
[  748.268035] posix_acl_update_mode: 0(3001) 1
[  748.268570] setfacl --set u::rw,g::r,o::r,m:rw test-file
[  748.274193] posix_acl_update_mode: 0(3001) 1
[  748.279198] capsh --drop=cap_fsetid -- -c setfacl --set u::rw,g::r,o::r test-file
[  748.287894] posix_acl_update_mode: 0(3001) 0

passing mnt_userns:
[   81.159766] setfacl --set u::rw,g::r,o::r,m:r test-file
[   81.165207] posix_acl_update_mode: 0(3001) 1
[   81.165226] posix_acl_update_mode: 0(3001) 1
[   81.165732] setfacl --set u::rw,g::r,o::r,m:rw test-file
[   81.170978] posix_acl_update_mode: 0(3001) 1
[   81.176014] capsh --drop=cap_fsetid -- -c setfacl --set u::rw,g::r,o::r test-file
[   81.184648] posix_acl_update_mode: 1(0) 0
[   81.184663] posix_acl_update_mode: 1(0) 0

The difference is that, by passing mnt_userns, the caller (root) is
in the owning group of the file (3001 -> 0). Without passing mnt_userns,
it is not (3001 -> 3001).

Christian, Andreas, could you confirm that this is correct?

Hey Robert,

s/Robert/Roberto/

Sorry for the typo.