From: Willy Tarreau <w@1wt.eu>
Date: Wed, 25 Oct 2023 13:02:59 +0000 (+0200)
Subject: DEBUG: lua: add tainted flags for stuck Lua contexts
X-Git-Tag: v2.9-dev9~29
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=26a6481f007c0f8a264c6d9bca891a67780edb9b;p=thirdparty%2Fhaproxy.git

DEBUG: lua: add tainted flags for stuck Lua contexts

William suggested that since we can detect the presence of Lua in the
stack, let's combine it with stuck detection to set a new pair of flags
indicating a stuck Lua context and a stuck Lua shared context.

Now, executing an infinite loop in a Lua sample fetch function with
yield disabled crashes with tainted=0xe40 if loaded from a lua-load
statement, or tainted=0x640 from a lua-load-per-thread statement.

In addition, at the end of the panic dump, we can check if Lua was
seen stuck and emit recommendations about lua-load-per-thread and
the choice of dependencies depending on the presence of threads
and/or shared context.
---

diff --git a/include/haproxy/bug.h b/include/haproxy/bug.h
index a716568f90..fbed1203d9 100644
--- a/include/haproxy/bug.h
+++ b/include/haproxy/bug.h
@@ -249,6 +249,8 @@ enum tainted_flags {
 	TAINTED_REDEFINITION           = 0x00000080, /* symbol redefinition detected */
 	TAINTED_REPLACED_MEM_ALLOCATOR = 0x00000100, /* memory allocator was replaced using LD_PRELOAD */
 	TAINTED_PANIC                  = 0x00000200, /* a panic dump has started */
+	TAINTED_LUA_STUCK              = 0x00000400, /* stuck in a Lua context */
+	TAINTED_LUA_STUCK_SHARED       = 0x00000800, /* stuck in a shared Lua context */
 };
 
 /* this is a bit field made of TAINTED_*, and is declared in haproxy.c */
diff --git a/src/debug.c b/src/debug.c
index 5598d212d2..fbab47e2e2 100644
--- a/src/debug.c
+++ b/src/debug.c
@@ -227,6 +227,19 @@ void ha_thread_dump_one(int thr, int from_signal)
 	ha_task_dump(buf, th_ctx->current, "             ");
 
 	if (stuck && thr == tid) {
+#ifdef USE_LUA
+		if (th_ctx->current &&
+		    th_ctx->current->process == process_stream && th_ctx->current->context) {
+			const struct stream *s = (const struct stream *)th_ctx->current->context;
+			struct hlua *hlua = s ? s->hlua : NULL;
+
+			if (hlua && hlua->T) {
+				mark_tainted(TAINTED_LUA_STUCK);
+				if (hlua->state_id == 0)
+					mark_tainted(TAINTED_LUA_STUCK_SHARED);
+			}
+		}
+#endif
 		/* We only emit the backtrace for stuck threads in order not to
 		 * waste precious output buffer space with non-interesting data.
 		 * Please leave this as the last instruction in this function
@@ -437,6 +450,25 @@ void ha_panic()
 		chunk_reset(&trash);
 	}
 
+#ifdef USE_LUA
+	if (get_tainted() & TAINTED_LUA_STUCK_SHARED && global.nbthread > 1) {
+		chunk_printf(&trash,
+			     "### Note: at least one thread was stuck in a Lua context loaded using the\n"
+			     "          'lua-load' directive, which is known for causing heavy contention\n"
+			     "          when used with threads. Please consider using 'lua-load-per-thread'\n"
+			     "          instead if your code is safe to run in parallel on multiple threads.\n");
+		DISGUISE(write(2, trash.area, trash.data));
+	}
+	else if (get_tainted() & TAINTED_LUA_STUCK) {
+		chunk_printf(&trash,
+			     "### Note: at least one thread was stuck in a Lua context in a way that suggests\n"
+			     "          heavy processing inside a dependency or a long loop that can't yield.\n"
+			     "          Please make sure any external code you may rely on is safe for use in\n"
+			     "          an event-driven engine.\n");
+		DISGUISE(write(2, trash.area, trash.data));
+	}
+#endif
+
 	for (;;)
 		abort();
 }