From 2507597c428364e77da32464c14a21acb7f4b9f5 Mon Sep 17 00:00:00 2001
From: Harald Welte <laforge@gnumonks.org>
Date: Mon, 11 Mar 2019 17:33:17 +0100
Subject: [PATCH] bankd: Implement thread-safe SIGUSR1 talloc context reporting

If the main thread receives SIGUSR1, we dump its talloc report and
then signal all worker threads so each can dump their own talloc
report.

Change-Id: I89e7e22de5557376bd5a9625662d99ac0badf00c
---
 src/bankd.h      |  5 +++++
 src/bankd_main.c | 34 +++++++++++++++++++++++++++++++---
 2 files changed, 36 insertions(+), 3 deletions(-)

diff --git a/src/bankd.h b/src/bankd.h
index 5411968..0ca1984 100644
--- a/src/bankd.h
+++ b/src/bankd.h
@@ -61,6 +61,8 @@ struct bankd_worker {
 
 	/* thread of this worker. */
 	pthread_t thread;
+	/* top talloc context for this worker/thread */
+	void *tall_ctx;
 
 	/* File descriptor of the TCP connection to the remsim-client (modem) */
 	struct {
@@ -101,6 +103,9 @@ struct bankd {
 	/* list of slot mappings. only ever modified in main thread! */
 	struct slotmaps *slotmaps;
 
+	/* pthread ID of main thread */
+	pthread_t main;
+
 	/* list of bankd_workers. accessed/modified by multiple threads; protected by mutex */
 	struct llist_head workers;
 	pthread_mutex_t workers_mutex;
diff --git a/src/bankd_main.c b/src/bankd_main.c
index f0cb813..bcf5907 100644
--- a/src/bankd_main.c
+++ b/src/bankd_main.c
@@ -56,10 +56,13 @@
 
 /* signal indicates to worker thread that its map has been deleted */
 #define SIGMAPDEL	SIGRTMIN+1
+
+static void handle_sig_usr1(int sig);
 static void handle_sig_mapdel(int sig);
 
 __thread void *talloc_asn1_ctx;
 struct bankd *g_bankd;
+static void *g_tall_ctx;
 
 static void *worker_main(void *arg);
 
@@ -71,7 +74,7 @@ int asn_debug;
 
 static void bankd_init(struct bankd *bankd)
 {
-	void *g_tall_ctx = talloc_named_const(NULL, 0, "global");
+	g_tall_ctx = talloc_named_const(NULL, 0, "global");
 	osmo_init_logging2(g_tall_ctx, &log_info);
 
 	asn_debug = 0;
@@ -241,7 +244,9 @@ int main(int argc, char **argv)
 
 	handle_options(argc, argv);
 
+	g_bankd->main = pthread_self();
 	signal(SIGMAPDEL, handle_sig_mapdel);
+	signal(SIGUSR1, handle_sig_usr1);
 
 	/* Connection towards remsim-server */
 	rc = server_conn_fsm_alloc(g_bankd, srvc);
@@ -330,6 +335,29 @@ static void handle_sig_mapdel(int sig)
 	worker_set_state(g_worker, BW_ST_CONN_CLIENT_UNMAPPED);
 }
 
+static void handle_sig_usr1(int sig)
+{
+	OSMO_ASSERT(sig == SIGUSR1);
+
+	if (pthread_equal(g_bankd->main, pthread_self())) {
+		struct bankd_worker *worker;
+		/* main thread */
+		fprintf(stderr, "=== Talloc Report of main thread:\n");
+		talloc_report(g_tall_ctx, stderr);
+
+		/* iterate over worker threads and ask them to dump their talloc state */
+		pthread_mutex_lock(&g_bankd->workers_mutex);
+		llist_for_each_entry(worker, &g_bankd->workers, list) {
+			pthread_kill(worker->thread, SIGUSR1);
+		}
+		pthread_mutex_unlock(&g_bankd->workers_mutex);
+	} else {
+		/* worker thread */
+		fprintf(stderr, "=== Talloc Report of %s\n", g_worker->name);
+		talloc_report(g_worker->tall_ctx, stderr);
+	}
+}
+
 static void worker_cleanup(void *arg)
 {
 	struct bankd_worker *worker = (struct bankd_worker *) arg;
@@ -705,8 +733,8 @@ static void *worker_main(void *arg)
 
 	/* not permitted in multithreaded environment */
 	talloc_disable_null_tracking();
-	top_ctx = talloc_named_const(NULL, 0, "top");
-	talloc_asn1_ctx = talloc_named_const(top_ctx, 0, "asn1");
+	g_worker->tall_ctx = talloc_named_const(NULL, 0, "top");
+	talloc_asn1_ctx = talloc_named_const(g_worker->tall_ctx, 0, "asn1");
 
 	/* set the thread name */
 	g_worker->name = talloc_asprintf(g_worker->tall_ctx, "bankd-worker(%u)", g_worker->num);