Redesign sidecar using byte-exact guest filenames without a sidecar index

.ci/check-matrix-lists.sh

@@ -0,0 +1,71 @@
+#!/usr/bin/env bash
+
+# Keep the test-matrix skip lists honest.
+#
+# tests/test-matrix.sh runs each test under two runners and carries a skip list
+# per runner: QEMU_SKIP for tests the reference kernel cannot adjudicate,
+# ELFUSE_SKIP for tests that need a writable byte-exact root the elfuse lane
+# does not have. Both lists are matched against a test's label by string, and a
+# string that matches nothing fails silently: the suite still reports success,
+# having run one test fewer than the reader believes.
+#
+# Two ways that goes wrong, both of which cost coverage without costing a red
+# build, and neither of which the suite itself can notice:
+#
+#   1. A label in a skip list that no longer names a registered test. Dead
+#      config: it reads as deliberate coverage policy while guarding nothing,
+#      and it hides the rename that orphaned it.
+#   2. A label in both lists at once. The test is then skipped under every
+#      runner the matrix has, so it never executes anywhere while still looking
+#      registered.
+#
+# The pass counts themselves are not checked here. test-matrix.sh already holds
+# each lane to its EXPECTED_BASELINES floor at runtime, which is a stronger
+# check than anything static, and duplicating it would only add a second number
+# to keep in step.
+
+set -e -u -o pipefail
+
+MATRIX="${1:-$(dirname "$0")/../tests/test-matrix.sh}"
+
+if [ ! -r "$MATRIX" ]; then
+    echo "check-matrix-lists: cannot read $MATRIX" >&2
+    exit 2
+fi
+
+# Body of a NAME="..." block spanning lines, one entry per line.
+list_entries()
+{
+    sed -n "/^$1=\"/,/^\"\$/p" "$MATRIX" | sed '1d;$d' | tr -s ' \t' '\n' \
+        | grep -v '^$' || true
+}
+
+# Labels registered with the test_* wrappers: the argument after "$runner".
+registered_labels()
+{
+    grep -oE '\btest_(check|rc|pipe) +"\$runner" +"[^"]+"' "$MATRIX" \
+        | sed -E 's/.*"\$runner" +"([^"]+)"$/\1/' | sort -u
+}
+
+ret=0
+registered="$(registered_labels)"
+
+for list in QEMU_SKIP ELFUSE_SKIP; do
+    while IFS= read -r label; do
+        [ -n "$label" ] || continue
+        if ! printf '%s\n' "$registered" | grep -qxF "$label"; then
+            echo "Error: $list names '$label', which no test_* call registers" >&2
+            ret=1
+        fi
+    done < <(list_entries "$list")
+done
+
+while IFS= read -r label; do
+    [ -n "$label" ] || continue
+    if list_entries QEMU_SKIP | grep -qxF "$label"; then
+        echo "Error: '$label' is in both QEMU_SKIP and ELFUSE_SKIP, so it never runs" >&2
+        ret=1
+    fi
+done < <(list_entries ELFUSE_SKIP)
+
+exit $ret

Makefile

@@ -39,7 +39,8 @@ SRCS := \
     syscall/mem.c \
     syscall/path.c \
     syscall/fuse.c \
-    syscall/sidecar.c \
+    syscall/casefold.c \
+    syscall/casefold-walk.c \
     syscall/chown-overlay.c \
     syscall/fs.c \
     syscall/fs-stat.c \
@@ -192,6 +193,29 @@ $(BUILD_DIR)/test-teardown-live-vcpu-host: \
 	@echo "  LD      $@"
 	$(Q)$(CC) $(CFLAGS) -o $@ $^ $(HVF_LDFLAGS)
 
+## Build the volume naming probe (native macOS binary)
+# Standalone: it measures the filesystem, so it links nothing from the project.
+$(BUILD_DIR)/probe-volume-naming: $(BUILD_DIR)/probe-volume-naming.o \
+		| $(BUILD_DIR)
+	@echo "  LD      $@"
+	$(Q)$(CC) $(CFLAGS) -o $@ $^
+
+## Build the filename codec host test (native macOS binary)
+# casefold.o is a leaf translation unit with no syscall-layer dependencies, so
+# the test links exactly the code under test and nothing else.
+$(BUILD_DIR)/test-casefold-host: $(BUILD_DIR)/test-casefold-host.o \
+		$(BUILD_DIR)/syscall/casefold.o | $(BUILD_DIR)
+	@echo "  LD      $@"
+	$(Q)$(CC) $(CFLAGS) -o $@ $^
+
+## Build the case-exact path resolution host test (native macOS binary)
+# Links the resolver and the codec; the two process-state symbols the resolver
+# reads are stubbed in the test.
+$(BUILD_DIR)/test-casefold-walk-host: $(BUILD_DIR)/test-casefold-walk-host.o \
+		$(BUILD_DIR)/syscall/casefold-walk.o \
+		$(BUILD_DIR)/syscall/casefold.o | $(BUILD_DIR)
+	@echo "  LD      $@"
+	$(Q)$(CC) $(CFLAGS) -o $@ $^
 
 # Guest test binaries (cross-compiled, aarch64-linux)
 # Only used when GUEST_TEST_BINARIES is not set.
@@ -214,6 +238,11 @@ $(BUILD_DIR)/test-pthread: tests/test-pthread.c | $(BUILD_DIR)
 	@echo "  CROSS   $< (with -lpthread)"
 	$(Q)$(CROSS_COMPILE)gcc -D_GNU_SOURCE -static -O2 -o $@ $< -lpthread
 
+# test-sysroot-name-soak churns from worker threads plus forked children
+$(BUILD_DIR)/test-sysroot-name-soak: tests/test-sysroot-name-soak.c | $(BUILD_DIR)
+	@echo "  CROSS   $< (with -lpthread)"
+	$(Q)$(CROSS_COMPILE)gcc -D_GNU_SOURCE -static -O2 -o $@ $< -lpthread
+
 # test-process-lifecycle creates a worker to verify that process PIDs and
 # thread TIDs share one namespace-wide allocator across fork children.
 $(BUILD_DIR)/test-process-lifecycle: tests/test-process-lifecycle.c src/utils.h | $(BUILD_DIR)

README.md

@@ -27,6 +27,9 @@ linker resolved against an external sysroot via `--sysroot`.
   signals, timers, futexes (incl. PI ops), and polling
 - Guest reads and writes the macOS filesystem directly; no overlay or
   volume mount layer
+- Linux byte-exact filename semantics under `--sysroot`, including
+  case-colliding names on the default case-folding APFS (see
+  [docs/filenames.md](docs/filenames.md))
 - Synthetic `/proc` and selected `/dev` emulation for user-space probes
 - Guest-internal FUSE: `/dev/fuse` and `mount("fuse")` work without
   macFUSE / FUSE-T / FSKit
@@ -53,7 +56,7 @@ boot-time overhead those tools impose.
 - GNU `objcopy` or `llvm-objcopy`
 - Hypervisor entitlement: `com.apple.security.hypervisor`
 
-To build only (`make elfuse`) without running tests, just the 
+To build only (`make elfuse`) without running tests, just the
 Xcode Command Line Tools and `objcopy` (`brew install binutils`) suffice.
 
 For guest test binaries, the project also expects an AArch64 Linux cross
@@ -62,7 +65,7 @@ used by the repository test harness, but `CROSS_COMPILE` and
 `BAREMETAL_CROSS` are overridable.
 
 See
-[docs/testing.md](docs/testing.md#build-requirements) for toolchain setup guide. 
+[docs/testing.md](docs/testing.md#build-requirements) for toolchain setup guide.
 
 ## Quick Start
 
@@ -109,6 +112,9 @@ The build signs `build/elfuse` before use. Override the signing identity with
 - [docs/testing.md](docs/testing.md): build prerequisites, the
   `make check` flow, the QEMU and Rosetta cross-check matrices, and
   fixture handling.
+- [docs/filenames.md](docs/filenames.md): how a guest filename becomes a
+  name on disk and back: case folding and normalization on the sysroot
+  volume, the escape encoding, and the length limits both systems impose.
 - [docs/internals.md](docs/internals.md): canonical technical
   reference -- runtime lifecycle, HVF constraints, EL1 shim and HVC
   protocol, page-table splitting, syscall translation tables, threads