Fix phpGH-13437: FPM: ERROR: scoreboard: failed to lock (already locked)

sapi/fpm/fpm/fpm_atomic.h

@@ -156,6 +156,25 @@ static inline int fpm_spinlock(atomic_t *lock, int try_once) /* {{{ */
 }
 /* }}} */
 
+static inline int fpm_spinlock_with_max_retries(atomic_t *lock, unsigned int max_retries)
+{
+    unsigned int retries = 0;
+
+    for (;;) {
+        if (atomic_cmp_set(lock, 0, 1)) {
+            return 1;
+        }
+
+        sched_yield();
+
+        if (++retries > max_retries) {
+            return 0;
+        }
+    }
+
+    return 1;
+}
+
 #define fpm_unlock(lock) lock = 0
 
 #endif

sapi/fpm/fpm/fpm_scoreboard.c

@@ -73,6 +73,24 @@ int fpm_scoreboard_init_main(void)
 	return 0;
 }
 
+static inline void fpm_scoreboard_readers_add(struct fpm_scoreboard_s *scoreboard, int num)
+{
+	/* this should be replaced with atomic_fetch_add but it is not currently supported on all platforms */
+	fpm_spinlock(&scoreboard->lock, 1);
+	scoreboard->reader_count += num;
+	fpm_unlock(scoreboard->lock);
+}
+
+static inline void fpm_scoreboard_readers_increment(struct fpm_scoreboard_s *scoreboard)
+{
+	fpm_scoreboard_readers_add(scoreboard, 1);
+}
+
+static inline void fpm_scoreboard_readers_decrement(struct fpm_scoreboard_s *scoreboard)
+{
+	fpm_scoreboard_readers_add(scoreboard, -1);
+}
+
 static struct fpm_scoreboard_s *fpm_scoreboard_get_for_update(struct fpm_scoreboard_s *scoreboard) /* {{{ */
 {
 	if (!scoreboard) {
@@ -93,7 +111,23 @@ void fpm_scoreboard_update_begin(struct fpm_scoreboard_s *scoreboard) /* {{{ */
 		return;
 	}
 
-	fpm_spinlock(&scoreboard->lock, 0);
+	if (!fpm_spinlock_with_max_retries(&scoreboard->writer_active, FPM_SCOREBOARD_SPINLOCK_MAX_RETRIES)) {
+		/* in this case the writer might have crashed so just warn and continue as the lock was acquired */
+		zlog(ZLOG_WARNING, "Writer waited too long for another writer to release lock.");
+	}
+
+	/* wait until no readers are active or lock is acquired */
+	int retries = 0;
+	while (scoreboard->reader_count > 0) {
+		if (++retries > FPM_SCOREBOARD_SPINLOCK_MAX_RETRIES) {
+			/* decrement reader count by 1 (assuming a killed or crashed reader) */
+			fpm_scoreboard_readers_decrement(scoreboard);
+			zlog(ZLOG_WARNING, "Writer detected a potential crashed reader, decrementing reader count.");
+			retries = 0;
+		}
+
+		sched_yield();
+	}
 }
 /* }}} */
 
@@ -170,7 +204,7 @@ void fpm_scoreboard_update_commit(
 		scoreboard->active_max = scoreboard->active;
 	}
 
-	fpm_unlock(scoreboard->lock);
+	fpm_unlock(scoreboard->writer_active);
 }
 /* }}} */
 
@@ -234,16 +268,25 @@ struct fpm_scoreboard_proc_s *fpm_scoreboard_proc_get_from_child(struct fpm_chil
 
 struct fpm_scoreboard_s *fpm_scoreboard_acquire(struct fpm_scoreboard_s *scoreboard, int nohang) /* {{{ */
 {
-	struct fpm_scoreboard_s *s;
-
-	s = scoreboard ? scoreboard : fpm_scoreboard;
+	struct fpm_scoreboard_s *s = scoreboard ? scoreboard : fpm_scoreboard;
 	if (!s) {
 		return NULL;
 	}
 
-	if (!fpm_spinlock(&s->lock, nohang)) {
-		return NULL;
+	fpm_scoreboard_readers_increment(s);
+
+	/* wait until no writer is active */
+	int retries = 0;
+	while (s->writer_active) {
+		sched_yield();
+
+        if (++retries > FPM_SCOREBOARD_SPINLOCK_MAX_RETRIES) {
+            zlog(ZLOG_WARNING, "Reader waited too long for writer to release lock.");
+			fpm_scoreboard_readers_decrement(s);
+			return NULL;
+        }
 	}
+
 	return s;
 }
 /* }}} */
@@ -253,7 +296,7 @@ void fpm_scoreboard_release(struct fpm_scoreboard_s *scoreboard) {
 		return;
 	}
 
-	scoreboard->lock = 0;
+	fpm_scoreboard_readers_decrement(scoreboard);
 }
 
 struct fpm_scoreboard_s *fpm_scoreboard_copy(struct fpm_scoreboard_s *scoreboard, int copy_procs)

sapi/fpm/fpm/fpm_scoreboard.h

@@ -18,6 +18,8 @@
 #define FPM_SCOREBOARD_LOCK_HANG 0
 #define FPM_SCOREBOARD_LOCK_NOHANG 1
 
+#define FPM_SCOREBOARD_SPINLOCK_MAX_RETRIES 50000
+
 struct fpm_scoreboard_proc_s {
 	union {
 		atomic_t lock;
@@ -52,6 +54,8 @@ struct fpm_scoreboard_s {
 		atomic_t lock;
 		char dummy[16];
 	};
+	atomic_t writer_active;
+	atomic_t reader_count;
 	char pool[32];
 	int pm;
 	time_t start_epoch;