libink/dbus: give parked and outstanding calls a deadline

A call is parked until the resolver says who sent it, and an outbound
call sits in a pending slot until its reply lands.  Neither had a way
to give up.  A broker that answers GetConnectionUnixUser slowly, or
not at all, leaves the caller waiting forever and keeps the slot; four
of those and every later privileged call is refused with
LimitsExceeded until Finit restarts.

libink cannot time itself out, it has no event loop, so the deadline
is the embedder's to keep.  One sweep per connection covers both, and
the ordering between them stays in the library rather than in each
embedder: calls first, because one timing out usually resolves the
park it was made for, and AccessDenied tells that caller more than a
bare timeout.

The sweep is armed when a resolve is deferred and stops rearming as
soon as nothing is outstanding, so a system that never meets a broker
never wakes up for it.

Signed-off-by: Joachim Wiberg <troglobit@gmail.com>

libink/connection.c

@@ -87,6 +87,7 @@ int link_connection_call(link_connection_t *conn, const char *destination,
 
 	bus->pending[i].used     = 1;
 	bus->pending[i].serial   = serial;
+	bus->pending[i].stamp    = __now_ms();
 	bus->pending[i].cb       = cb;
 	bus->pending[i].userdata = userdata;
 
@@ -107,6 +108,45 @@ void __bus_free(link_connection_t *conn)
 	conn->bus = NULL;
 }
 
+/* Neither half can time itself out, libink has no event loop, so the
+ * embedder sweeps.  Calls go first: one timing out usually resolves
+ * the park it was made for, and being told AccessDenied says more to
+ * that caller than a bare timeout. */
+int link_connection_expire(link_connection_t *conn, unsigned int age_ms)
+{
+	struct link_bus *bus;
+	uint64_t now;
+	int i, live = 0;
+
+	if (!conn || !conn->bus)
+		return 0;
+
+	bus = conn->bus;
+	now = __now_ms();
+	for (i = 0; i < LINK_PENDING_CAP; i++) {
+		link_reply_cb_t  cb;
+		void            *userdata;
+
+		if (!bus->pending[i].used)
+			continue;
+
+		if (now - bus->pending[i].stamp < age_ms) {
+			live++;
+			continue;
+		}
+
+		cb       = bus->pending[i].cb;
+		userdata = bus->pending[i].userdata;
+		bus->pending[i].used = 0;
+
+		__dbg("timed out call serial %u, no reply", bus->pending[i].serial);
+		if (cb)
+			cb(conn, NULL, userdata);
+	}
+
+	return live + __dispatch_expire_parked(conn, age_ms);
+}
+
 void link_connection_close(link_connection_t *conn)
 {
 	size_t i;

libink/dispatch.c

@@ -387,9 +387,10 @@ static struct link_parked *park(link_connection_t *conn, const uint8_t *frame,
 	if (i == LINK_PARKED_CAP)
 		return NULL;
 
-	bus->parked[i].tok  = ++bus->next_tok;
-	bus->parked[i].conn = conn;
-	bus->parked[i].len  = len;
+	bus->parked[i].tok   = ++bus->next_tok;
+	bus->parked[i].conn  = conn;
+	bus->parked[i].stamp = __now_ms();
+	bus->parked[i].len   = len;
 	memcpy(bus->parked[i].buf, frame, len);
 	*tok = bus->parked[i].tok;
 
@@ -402,6 +403,51 @@ static void unpark(struct link_parked *p)
 	p->conn = NULL;
 }
 
+/* A resolver that answers late, or never, would otherwise hold both
+ * the slot and the caller forever.  Returns how many are still parked;
+ * link_connection_expire() is what calls this. */
+int __dispatch_expire_parked(link_connection_t *conn, unsigned int age_ms)
+{
+	uint64_t now;
+	int i, live = 0;
+
+	if (!conn->bus)
+		return 0;
+
+	now = __now_ms();
+	for (i = 0; i < LINK_PARKED_CAP; i++) {
+		struct link_parked *p = &conn->bus->parked[i];
+		uint8_t             buf[LINK_PARKED_MSG_MAX];
+		struct link_msg     msg;
+		size_t              len;
+
+		if (!p->tok)
+			continue;
+
+		if (now - p->stamp < age_ms) {
+			live++;
+			continue;
+		}
+
+		/* Free the slot before replying, as link_uid_resolved()
+		 * does: the send path must not find it still parked. */
+		len = p->len;
+		memcpy(buf, p->buf, len);
+		unpark(p);
+
+		if (__msg_parse(buf, len, &msg) <= 0)
+			continue;
+
+		__dbg("timed out %s, nobody said who the caller was",
+		      msg.member ? msg.member : "call");
+		(void)__send_error(conn, &msg,
+				   "org.freedesktop.DBus.Error.TimedOut",
+				   "Timed out identifying the caller");
+	}
+
+	return live;
+}
+
 void __dispatch_forget_conn(link_connection_t *conn)
 {
 	struct link_bus *bus = conn->bus;

libink/internal.h

@@ -62,10 +62,12 @@ TAILQ_HEAD(link_object_list, link_object);
 /* An inbound method call held while we find out who sent it.  The
  * message is copied because rxbuf is reused as soon as we return to
  * the read loop.  `tok` is the handle the resolver answers with, and
- * zero when the slot is free. */
+ * zero when the slot is free.  `stamp` is when it was parked, for
+ * link_connection_expire(). */
 struct link_parked {
 	link_authz_t        tok;
 	link_connection_t  *conn;
+	uint64_t            stamp;
 	size_t              len;
 	uint8_t             buf[LINK_PARKED_MSG_MAX];
 };
@@ -86,6 +88,7 @@ struct link_bus {
 	struct {
 		int              used;
 		uint32_t         serial;
+		uint64_t         stamp;	/* for link_connection_expire() */
 		link_reply_cb_t  cb;
 		void            *userdata;
 	} pending[LINK_PENDING_CAP];
@@ -177,9 +180,11 @@ void __log(const char *func, const char *fmt, ...)
 	__attribute__((format(printf, 2, 3)));
 #define __dbg(fmt, ...)  __log(__func__, fmt, ##__VA_ARGS__)
 
-/* io.c — shared EINTR-resilient I/O loops. */
+/* io.c — shared EINTR-resilient I/O loops, and the clock the expiry
+ * sweeps measure against. */
 int  __io_write_all(int fd, const void *buf, size_t len);
 int  __io_read_full(int fd, void *buf,       size_t len);
+uint64_t __now_ms(void);
 
 /* auth.c */
 int  __auth_process(link_connection_t *conn);
@@ -193,6 +198,7 @@ void             __bus_free(link_connection_t *conn);
 /* dispatch.c */
 int  __dispatch_message(link_connection_t *conn, const struct link_msg *m, size_t framelen);
 void __dispatch_forget_conn(link_connection_t *conn);
+int  __dispatch_expire_parked(link_connection_t *conn, unsigned int age_ms);
 int  __send_error(link_connection_t *conn, const struct link_msg *req,
 		     const char *error_name, const char *text);
 int  __send_method_return(link_connection_t *conn, const struct link_msg *req,

libink/io.c

@@ -5,15 +5,30 @@
  * read_full.  On any other error they return -1 with an unknown
  * number of bytes already transferred.
  *
+ * Also the clock the expiry sweeps measure against: monotonic, so a
+ * step in wall time cannot make a call look older or younger than it
+ * is.
+ *
  * Copyright (c) 2026  Joachim Wiberg <troglobit@gmail.com>
  * SPDX-License-Identifier: MIT
  */
 
 #include <errno.h>
+#include <time.h>
 #include <unistd.h>
 
 #include "internal.h"
 
+uint64_t __now_ms(void)
+{
+	struct timespec ts;
+
+	if (clock_gettime(CLOCK_MONOTONIC, &ts))
+		return 0;
+
+	return (uint64_t)ts.tv_sec * 1000 + (uint64_t)(ts.tv_nsec / 1000000);
+}
+
 int __io_write_all(int fd, const void *buf, size_t len)
 {
 	const char *p = buf;

libink/link.h

@@ -164,6 +164,18 @@ int link_connection_call(link_connection_t *conn, const char *destination,
 			 link_reply_cb_t cb, void *userdata,
 			 const char *signature, ...);
 
+/* Nothing in libink runs a clock, it has no event loop, so calls that
+ * go unanswered in either direction are the embedder's to time out.
+ * This drops anything held longer than `age_ms` on one connection: a
+ * park whose resolver never answered, which leaves its caller
+ * org.freedesktop.DBus.Error.TimedOut, and a call whose reply never
+ * came, whose callback runs once with a NULL reply exactly as a
+ * dropped connection would.
+ *
+ * Returns how many are still outstanding, so a sweep can stop
+ * rearming once nothing is left. */
+int link_connection_expire(link_connection_t *conn, unsigned int age_ms);
+
 /* ----------  server / connection lifecycle  ---------- */
 
 /* Bind a listening socket at `path` with file mode `mode`, e.g. 0660