Add comment for how O_CREAT and O_TRUNC is handled
Bug: 205172873
Test: Wrote a small program to verify the theory
Change-Id: Ic89f10884b2f25d86cfe027ea3541e5a5a5ad621
diff --git a/authfs/fd_server/src/aidl.rs b/authfs/fd_server/src/aidl.rs
index 125b991..f13f249 100644
--- a/authfs/fd_server/src/aidl.rs
+++ b/authfs/fd_server/src/aidl.rs
@@ -290,8 +290,10 @@
let new_fd = openat(
dir.as_raw_fd(),
basename,
- // TODO(205172873): handle the case when the file already exist, e.g. truncate
- // or fail, and possibly allow the client to specify. For now, always truncate.
+ // This function is supposed to be only called when FUSE/authfs thinks the file
+ // does not exist. However, if the file does exist from the view of fd_server
+ // (where the execution context is considered untrusted), we prefer to honor
+ // authfs and still allow the create to success. Therefore, always use O_TRUNC.
OFlag::O_CREAT | OFlag::O_RDWR | OFlag::O_TRUNC,
mode,
)
diff --git a/authfs/src/fusefs.rs b/authfs/src/fusefs.rs
index 84129b6..9c17fc4 100644
--- a/authfs/src/fusefs.rs
+++ b/authfs/src/fusefs.rs
@@ -664,7 +664,6 @@
_flags: u32,
umask: u32,
) -> io::Result<(Entry, Option<Self::Handle>, FuseOpenOptions)> {
- // TODO(205172873): handle O_TRUNC and O_EXCL properly.
let new_inode = self.create_new_entry_with_ref_count(
parent,
name,