Skip to contentSkip to conversation

Transmute flag is not inheritted on overlay fs

RFC PATCH New Mengchi Cheng 2 participants 4 messages Apr 19, 2023

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

Hello,

On the overlay ext4 file system, we found that transmute flag is not inherited by newly created sub-directories. The issue can be recreated on the newest kernel(6.3.0-rc6) on qemux86-64 with following steps.

/data directory is mounted on /dev/vdb which is a ext4 fs. It is remounted as an overlay again to upperdir /home/root/data.

# mount -t overlay overlay -o lowerdir=/data,upperdir=/home/root/data,workdir=/home/root/data_work /data
Add a new smack rule and set label and flag to /data directory.
# echo "_ system rwxatl" > /sys/fs/smackfs/load2
# chsmack -a "system" /data
# chsmack -t /data
Create directories under /data.
# mkdir -p /data/dir1/dir2
And then check the smack label of dir1 and dir2.
# chsmack /data/dir1
/data/dir1 access="system"
# chsmack /data/dir1/dir2
/data/dir1/dir2 access="_"
We can see dir1 did not inherit transmute flag from data and dir2 got the
process label.

The transmute xattr of the inode is set inside the smack_d_instantiate which depends on SMK_INODE_CHANGED bit of isp->smk_flags. But the bit is not set in the overlay fs mkdir function call chain. So one simple solution we have is passing inode ptr into smack_dentry_create_files_as and set the SMK_INODE_CHANGED bit if parent dir is transmuting. Although it looks reasonable to me and we did not meet any issue in testing, I am not sure if there is a better solution to it. It will be great, if experts could take a look.

Thanks,
Mengchi Cheng

maintainer
#1 lore raw
Quoted 23 lines · Mengchi Cheng

Hello,

On the overlay ext4 file system, we found that transmute flag is not inherited by newly created sub-directories. The issue can be recreated on the newest kernel(6.3.0-rc6) on qemux86-64 with following steps.

/data directory is mounted on /dev/vdb which is a ext4 fs. It is remounted as an overlay again to upperdir /home/root/data.

# mount -t overlay overlay -o lowerdir=/data,upperdir=/home/root/data,workdir=/home/root/data_work /data
Add a new smack rule and set label and flag to /data directory.
# echo "_ system rwxatl" > /sys/fs/smackfs/load2
# chsmack -a "system" /data
# chsmack -t /data
Create directories under /data.
# mkdir -p /data/dir1/dir2
And then check the smack label of dir1 and dir2.
# chsmack /data/dir1
/data/dir1 access="system"
# chsmack /data/dir1/dir2
/data/dir1/dir2 access="_"
We can see dir1 did not inherit transmute flag from data and dir2 got the
process label.
Quoted 8 lines

The transmute xattr of the inode is set inside the smack_d_instantiate which depends on SMK_INODE_CHANGED bit of isp->smk_flags. But the bit is not set in the overlay fs mkdir function call chain. So one simple solution we have is passing inode ptr into smack_dentry_create_files_as and set the SMK_INODE_CHANGED bit if parent dir is transmuting. Although it looks reasonable to me and we did not meet any issue in testing, I am not sure if there is a better solution to it. It will be great, if experts could take a look.

I will be happy to look at your solution. Please post a patch.

Quoted 5 lines

Thanks,
Mengchi Cheng

#2 lore raw

RFC PATCH 1/1 Set SMK_INODE_CHANGED inside smack_dentry_create_files_as

On the overlay fs with smack lsm enabled, new subdir did not inherit transmute xattr form parent dir.

One solution that can solve it is passing inode into smack_dentry_create_files_as. And Set SMK_INODE_CHANGED to smak_flags if directory has transmute xattr.

---
 fs/overlayfs/dir.c            | 2 +-
 include/linux/lsm_hook_defs.h | 2 +-
 include/linux/security.h      | 4 ++--
 security/security.c           | 4 ++--
 security/selinux/hooks.c      | 2 +-
 security/smack/smack_lsm.c    | 8 ++++++--
 6 files changed, 13 insertions(+), 9 deletions(-)
fs/overlayfs/dir.c +1 -1
index fc25fb95d5fc..1b3f7f3a5468 100644
@@ -598,7 +598,7 @@ static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 		override_cred->fsgid = inode->i_gid;
 		err = security_dentry_create_files_as(dentry,
 				attr->mode, &dentry->d_name, old_cred,
-				override_cred);
+				override_cred, inode);
 		if (err) {
 			put_cred(override_cred);
 			goto out_revert_creds;
include/linux/lsm_hook_defs.h +1 -1
index 094b76dc7164..96f1fdc21cbc 100644
@@ -84,7 +84,7 @@ LSM_HOOK(int, -EOPNOTSUPP, dentry_init_security, struct dentry *dentry,
 	 int mode, const struct qstr *name, const char **xattr_name,
 	 void **ctx, u32 *ctxlen)
 LSM_HOOK(int, 0, dentry_create_files_as, struct dentry *dentry, int mode,
-	 struct qstr *name, const struct cred *old, struct cred *new)
+	 struct qstr *name, const struct cred *old, struct cred *new, struct inode *inode)
 
 #ifdef CONFIG_SECURITY_PATH
 LSM_HOOK(int, 0, path_unlink, const struct path *dir, struct dentry *dentry)
include/linux/security.h +2 -2
index 5984d0d550b4..354d68dc69c5 100644
@@ -325,7 +325,7 @@ int security_dentry_init_security(struct dentry *dentry, int mode,
 int security_dentry_create_files_as(struct dentry *dentry, int mode,
 					struct qstr *name,
 					const struct cred *old,
-					struct cred *new);
+					struct cred *new, struct inode *inode);
 int security_path_notify(const struct path *path, u64 mask,
 					unsigned int obj_type);
 int security_inode_alloc(struct inode *inode);
@@ -756,7 +756,7 @@ static inline int security_dentry_init_security(struct dentry *dentry,
 static inline int security_dentry_create_files_as(struct dentry *dentry,
 						  int mode, struct qstr *name,
 						  const struct cred *old,
-						  struct cred *new)
+						  struct cred *new, struct inode *inode)
 {
 	return 0;
 }
security/security.c +2 -2
index cf6cc576736f..0ffe98cc57fe 100644
@@ -1110,10 +1110,10 @@ EXPORT_SYMBOL(security_dentry_init_security);
 
 int security_dentry_create_files_as(struct dentry *dentry, int mode,
 				    struct qstr *name,
-				    const struct cred *old, struct cred *new)
+				    const struct cred *old, struct cred *new, struct inode *inode)
 {
 	return call_int_hook(dentry_create_files_as, 0, dentry, mode,
-				name, old, new);
+				name, old, new, inode);
 }
 EXPORT_SYMBOL(security_dentry_create_files_as);
 
security/selinux/hooks.c +1 -1
index 9a5bdfc21314..2addc513bbb0 100644
@@ -2848,7 +2848,7 @@ static int selinux_dentry_init_security(struct dentry *dentry, int mode,
 static int selinux_dentry_create_files_as(struct dentry *dentry, int mode,
 					  struct qstr *name,
 					  const struct cred *old,
-					  struct cred *new)
+					  struct cred *new, struct inode *inode)
 {
 	u32 newsid;
 	int rc;
security/smack/smack_lsm.c +6 -2
index cfcbb748da25..e929e3e131c2 100644
@@ -4739,12 +4739,14 @@ static int smack_inode_copy_up_xattr(const char *name)
 static int smack_dentry_create_files_as(struct dentry *dentry, int mode,
 					struct qstr *name,
 					const struct cred *old,
-					struct cred *new)
+					struct cred *new,
+					struct inode *inode)
 {
 	struct task_smack *otsp = smack_cred(old);
 	struct task_smack *ntsp = smack_cred(new);
 	struct inode_smack *isp;
 	int may;
+	struct inode_smack *issp = smack_inode(inode);
 
 	/*
 	 * Use the process credential unless all of
@@ -4769,8 +4771,10 @@ static int smack_dentry_create_files_as(struct dentry *dentry, int mode,
 		 * providing access is transmuting use the containing
 		 * directory label instead of the process label.
 		 */
-		if (may > 0 && (may & MAY_TRANSMUTE))
+		if (may > 0 && (may & MAY_TRANSMUTE)) {
 			ntsp->smk_task = isp->smk_inode;
+			issp->smk_flags |= SMK_INODE_CHANGED;
+		}
 	}
 	return 0;
 }
Signature · 1 line
2.25.1
#3 lore raw
In reply to Casey Schaufler
Quoted 34 lines · Casey Schaufler
Quoted 23 lines · Mengchi Cheng

Hello,

On the overlay ext4 file system, we found that transmute flag is not inherited by newly created sub-directories. The issue can be recreated on the newest kernel(6.3.0-rc6) on qemux86-64 with following steps.

/data directory is mounted on /dev/vdb which is a ext4 fs. It is remounted as an overlay again to upperdir /home/root/data.

# mount -t overlay overlay -o lowerdir=/data,upperdir=/home/root/data,workdir=/home/root/data_work /data
Add a new smack rule and set label and flag to /data directory.
# echo "_ system rwxatl" > /sys/fs/smackfs/load2
# chsmack -a "system" /data
# chsmack -t /data
Create directories under /data.
# mkdir -p /data/dir1/dir2
And then check the smack label of dir1 and dir2.
# chsmack /data/dir1
/data/dir1 access="system"
# chsmack /data/dir1/dir2
/data/dir1/dir2 access="_"
We can see dir1 did not inherit transmute flag from data and dir2 got the
process label.
Quoted 8 lines

The transmute xattr of the inode is set inside the smack_d_instantiate which depends on SMK_INODE_CHANGED bit of isp->smk_flags. But the bit is not set in the overlay fs mkdir function call chain. So one simple solution we have is passing inode ptr into smack_dentry_create_files_as and set the SMK_INODE_CHANGED bit if parent dir is transmuting. Although it looks reasonable to me and we did not meet any issue in testing, I am not sure if there is a better solution to it. It will be great, if experts could take a look.

Quoted 2 lines

I will be happy to look at your solution. Please post a patch.

Sorry, it takes me a while to review and send out the patch. It contains a few files because it breaks kernel API. But the core is only in the change of smack_dentry_create_files_as.

If Roberto's patch will work, we can drop it. I posted my concern in that thread.
https://lore.kernel.org/all/[email protected]/

Quoted 6 lines
Quoted 5 lines

Thanks,
Mengchi Cheng