perf: add FOPEN_DIRECT_IO to bypass kernel page cache

Every read now goes through our FUSE handler instead of being served
from the kernel page cache. This gives accurate benchmark numbers for
CAS throughput and matches mountpoint-s3's behavior.

Also fix fs_bench.sh: add --uid/--gid for sudo mounts, use
create_on_open=1 for populate, and fail on mount errors.

Author: XciD

scripts/fs_bench.sh

@@ -128,6 +128,11 @@ do_mount() {
   )
   [[ "${HF_NO_DISK_CACHE:-0}" == "1" ]] && mount_args+=(--no-disk-cache)
   [[ "${HF_ADVANCED_WRITES:-0}" == "1" ]] && mount_args+=(--advanced-writes)
+  # When running as root (sudo), set file ownership to the calling user
+  # so fio and other tools can access mounted files without permission issues.
+  if [[ "$(id -u)" == "0" && -n "${SUDO_UID:-}" ]]; then
+    mount_args+=(--uid "${SUDO_UID}" --gid "${SUDO_GID:-${SUDO_UID}}")
+  fi
   mount_args+=("$@")
   mount_args+=(bucket "${HF_BENCH_BUCKET}" "${_MOUNT_DIR}")
 
@@ -138,7 +143,7 @@ do_mount() {
       { echo "Mount ready after $((i*500))ms" >&2; return 0; }
     sleep 0.5
   done
-  echo "Warning: mount not ready after 15s" >&2
+  echo "ERROR: mount not ready after 15s" >&2; exit 1
 }
 
 do_unmount() {
@@ -156,6 +161,10 @@ do_unmount() {
   _CHILD_PID=""; _MOUNT_DIR=""
 }
 
+should_run_job() {
+  [[ -z "${HF_JOB_NAME_FILTER:-}" ]] || [[ "$1" == *"${HF_JOB_NAME_FILTER}"* ]]
+}
+
 # ── File setup (one-time) ─────────────────────────────────────────────────────
 # Write all benchmark files through hf-mount's write path, flush to CAS, then unmount.
 # This mirrors mountpoint-s3's "create_only" step, but batched for efficiency.
@@ -298,11 +307,16 @@ make_write_job() {
   local read_job="$1"
   local write_job
   write_job=$(mktemp /tmp/fio-write-XXXX.fio)
+  # create_on_open=1: skip the layout step (create+ftruncate) and create at open time,
+  # so fio opens with O_CREAT|O_TRUNC which streaming write mode supports.
+  # Remove numjobs so populate writes a single copy of each file.
   sed -e 's/rw=\(rand\)\?read/rw=write/' \
       -e '/^time_based/d' \
       -e '/^runtime=/d' \
       -e '/^direct=/d' \
+      -e '/^numjobs=/d' \
       "${read_job}" > "${write_job}"
+  echo "create_on_open=1" >> "${write_job}"
   echo "${write_job}"
 }
 

src/virtual_fs/tests.rs

@@ -945,27 +945,6 @@ fn setattr_simple_mode_noop() {
     });
 }
 
-/// setattr(size) on an empty file in simple mode is a silent noop.
-#[test]
-fn setattr_simple_mode_empty_file_noop() {
-    let hub = MockHub::new();
-    let xet = MockXet::new();
-    let (rt, vfs) = vfs_simple(&hub, &xet);
-
-    rt.block_on(async {
-        let (attr, _fh) = vfs
-            .create(ROOT_INODE, "new.txt", 0o644, 1000, 1000, Some(1))
-            .await
-            .unwrap();
-        // ftruncate(fd, N) in simple mode succeeds but is a noop
-        let result = vfs.setattr(attr.ino, Some(100), None, None, None, None, None).await;
-        assert!(result.is_ok(), "ftruncate should succeed (noop): {:?}", result);
-        let inodes = vfs.inode_table.read().unwrap();
-        let entry = inodes.get(attr.ino).unwrap();
-        assert_eq!(entry.size, 0, "size should be unchanged (noop)");
-    });
-}
-
 /// setattr(size) on a directory returns EISDIR.
 #[test]
 fn setattr_directory_eisdir() {