Find the stale-pid test's registry by its members

mk/tests.mk

@@ -732,37 +732,43 @@ test-absock-cleanup: $(ELFUSE_BIN) $(BUILD_DIR)/test-absock-cleanup
 # pid to another elfuse process. The recipe plants such a record, host pid of
 # a live unrelated elfuse run with a start time it does not have, and the
 # family's kill(99, 0) must still fail with ESRCH.
+# Concurrent runners share the directory and a stale registry can carry a
+# reused pid, so select the registry that is new since the recipe started and
+# holds records for the family root and one of its children.
 ## registry ignores a reused host pid, and is not named after one
 test-registry-stale-pid: $(ELFUSE_BIN) $(BUILD_DIR)/test-registry-stale-pid
 	@tmp=$$(mktemp -d); xpid=; fpid=; \
 	trap 'kill $$xpid $$fpid 2>/dev/null; rm -rf "$$tmp"' EXIT; \
 	fail() { printf "FAIL: %s\n" "$$1"; exit 1; }; \
+	regs() { (cd "$$dir" && ls) | grep '^elfuse-procs-' | sort; }; \
 	printf "  %-30s " "stale record on reused pid"; \
 	tmo=$$(command -v timeout 2>/dev/null \
 	    || command -v gtimeout 2>/dev/null || true); \
 	[ -n "$$tmo" ] || { printf "SKIP (timeout(1) missing)\n"; exit 0; }; \
 	secs=$${TEST_TIMEOUT:-10}; \
 	dir=$$(getconf DARWIN_USER_TEMP_DIR); \
-	ls "$$dir" | grep '^elfuse-procs-' | sort > "$$tmp/before" || true; \
+	regs > "$$tmp/before" || true; \
 	mkfifo "$$tmp/go"; \
 	$$tmo "$$secs" $(ELFUSE_BIN) $(BUILD_DIR)/test-registry-stale-pid hold & \
 	xpid=$$!; \
 	$$tmo "$$secs" $(ELFUSE_BIN) $(BUILD_DIR)/test-registry-stale-pid \
 	    < "$$tmp/go" > "$$tmp/out" & \
 	fpid=$$!; \
 	exec 4> "$$tmp/go"; \
-	for i in $$(seq 1 50); do \
+	for i in $$(seq 1 $$((secs > 0 ? secs * 10 : 50))); do \
 		grep -q READY "$$tmp/out" && break; \
 		sleep 0.1; \
 	done; \
 	grep -q READY "$$tmp/out" || fail "family never reported READY"; \
-	new=$$(ls "$$dir" | grep '^elfuse-procs-' | sort \
-	    | comm -13 "$$tmp/before" -); \
-	[ "$$(printf '%s\n' "$$new" | grep -c .)" = 1 ] \
-	    || fail "expected one new registry in $$dir, saw '$$new'"; \
-	reg="$$dir$$new"; \
 	root=$$(pgrep -P "$$fpid" | head -1); \
 	[ -n "$$root" ] || fail "family root process not found"; \
+	kids=$$(pgrep -P "$$root" | paste -sd'|' -); \
+	[ -n "$$kids" ] || fail "family child process not found"; \
+	new=$$(regs | comm -13 "$$tmp/before" - | (cd "$$dir" \
+	    && xargs grep -ls "^$$root " | xargs grep -lsE "^($$kids) ")); \
+	[ "$$(printf '%s\n' "$$new" | grep -c .)" = 1 ] \
+	    || fail "expected one registry for host pids $$root/$$kids, saw '$$new'"; \
+	reg="$$dir$$new"; \
 	[ "$$new" != "elfuse-procs-$$root" ] \
 	    || fail "family id is the root pid $$root, which macOS recycles"; \
 	holder=$$(pgrep -P "$$xpid" | head -1); \

src/syscall/proc.c

@@ -400,17 +400,34 @@ int rseq_try_abort(guest_t *g,
     return result;
 }
 
-static bool process_pid_sequence_path(char *out, size_t out_size)
+/* Build "<dir>elfuse-<stem>-<id>" in the per-user private temp directory macOS
+ * provisions (mode 0700, owned by the invoking uid) rather than world-writable
+ * /tmp, so another local user cannot pre-plant a symlink or forge a file's
+ * contents. confstr returns the same directory for every process of this uid,
+ * so every family member agrees on the path. Callers open the result O_NOFOLLOW
+ * for defense in depth.
+ *
+ * Returns false (fail closed) if the directory cannot be resolved or the path
+ * does not fit.
+ */
+static bool private_tmp_path(char *out,
+                             size_t out_size,
+                             const char *stem,
+                             unsigned long long id)
 {
     char dir[PATH_MAX];
     size_t n = confstr(_CS_DARWIN_USER_TEMP_DIR, dir, sizeof(dir));
     if (n == 0 || n > sizeof(dir))
         return false;
-    int len = snprintf(out, out_size, "%selfuse-pidseq-%llu", dir,
-                       (unsigned long long) absock_get_namespace_id());
+    int len = snprintf(out, out_size, "%selfuse-%s-%llu", dir, stem, id);
     return len > 0 && (size_t) len < out_size;
 }
 
+static bool process_pid_sequence_path(char *out, size_t out_size)
+{
+    return private_tmp_path(out, out_size, "pidseq", absock_get_namespace_id());
+}
+
 int64_t proc_alloc_pid(void)
 {
     static _Atomic bool owner_sequence_reset;
@@ -587,36 +604,16 @@ pid_t proc_guest_to_host_pid(int64_t gpid)
     return result;
 }
 
-/* Build the path to the cross-process signal-transport file for @host_pid.
- *
- * Files live in the per-user private temp directory macOS provisions (mode
- * 0700, owned by the invoking uid) rather than world-writable /tmp, so another
- * local user cannot pre-plant a symlink to redirect the write or inject signal
- * numbers into the guest. confstr returns the same directory for every process
- * of this uid, so sender and receiver agree on the path. Callers open the
- * result O_NOFOLLOW for defense in depth.
- *
- * Returns false (fail closed) if the private directory cannot be resolved.
- */
+/* Cross-process signal-transport file for @host_pid, a live process. */
 static bool signal_transport_path(char *out, size_t out_size, pid_t host_pid)
 {
-    char dir[PATH_MAX];
-    size_t n = confstr(_CS_DARWIN_USER_TEMP_DIR, dir, sizeof(dir));
-    if (n == 0 || n > sizeof(dir))
-        return false;
-    int len = snprintf(out, out_size, "%selfuse-sig-%ld", dir, (long) host_pid);
-    return len > 0 && (size_t) len < out_size;
+    return private_tmp_path(out, out_size, "sig",
+                            (unsigned long long) host_pid);
 }
 
 static bool process_registry_path(char *out, size_t out_size)
 {
-    char dir[PATH_MAX];
-    size_t n = confstr(_CS_DARWIN_USER_TEMP_DIR, dir, sizeof(dir));
-    if (n == 0 || n > sizeof(dir))
-        return false;
-    int len = snprintf(out, out_size, "%selfuse-procs-%llu", dir,
-                       (unsigned long long) absock_get_namespace_id());
-    return len > 0 && (size_t) len < out_size;
+    return private_tmp_path(out, out_size, "procs", absock_get_namespace_id());
 }
 
 /* The process-group registry above intentionally contains only live host
@@ -654,13 +651,7 @@ typedef struct {
 
 static bool lifecycle_registry_path(char *out, size_t out_size)
 {
-    char dir[PATH_MAX];
-    size_t n = confstr(_CS_DARWIN_USER_TEMP_DIR, dir, sizeof(dir));
-    if (n == 0 || n > sizeof(dir))
-        return false;
-    int len = snprintf(out, out_size, "%selfuse-life-%llu", dir,
-                       (unsigned long long) absock_get_namespace_id());
-    return len > 0 && (size_t) len < out_size;
+    return private_tmp_path(out, out_size, "life", absock_get_namespace_id());
 }
 
 static bool lifecycle_registry_size(uint32_t capacity, size_t *size_out)