Fix security check for joint context= and fscontext= mount options

After some discussion on the actual meaning of the filesystem class security check in try context mount it was determined that the checks for the context= mount options were not correct if fscontext mount option had already been used. When labeling the superblock we should be checking relabel_from and relabel_to. But if the superblock has already been labeled (with fscontext) then context= is actually labeling the inodes, and so we should be checking relabel_from and associate. This patch fixes which checks are called depending on the mount options.

This is issue is in 2.6.8-rc1-git4 and should probably be fixed before 2.6.18 releases.

Signed-off-by Eric Paris <[email protected]>

 security/selinux/hooks.c |   14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

security/selinux/hooks.c +9 -5

@@ -523,12 +523,16 @@ static int try_context_mount(struct supe
 			goto out_free;
 		}
 
-		rc = may_context_mount_sb_relabel(sid, sbsec, tsec);
-		if (rc)
-			goto out_free;
-
-		if (!fscontext)
+		if (!fscontext) {
+			rc = may_context_mount_sb_relabel(sid, sbsec, tsec);
+			if (rc)
+				goto out_free;
 			sbsec->sid = sid;
+		} else {
+			rc = may_context_mount_inode_relabel(sid, sbsec, tsec);
+			if (rc)
+				goto out_free;
+		}
 		sbsec->mntpoint_sid = sid;
 
 		sbsec->behavior = SECURITY_FS_USE_MNTPOINT;

Quoted 9 lines · Eric Paris

After some discussion on the actual meaning of the filesystem class security check in try context mount it was determined that the checks for the context= mount options were not correct if fscontext mount option had already been used. When labeling the superblock we should be checking relabel_from and relabel_to. But if the superblock has already been labeled (with fscontext) then context= is actually labeling the inodes, and so we should be checking relabel_from and associate. This patch fixes which checks are called depending on the mount options.

Quoted 4 lines

This is issue is in 2.6.8-rc1-git4 and should probably be fixed before 2.6.18 releases.

Signed-off-by Eric Paris <[email protected]>

Acked-by Stephen Smalley <[email protected]>