From 8e127e75527a02c2b3854f94b419cf8138a378cf Mon Sep 17 00:00:00 2001
From: Costa Tsaousis <costa@netdata.cloud>
Date: Fri, 10 Jan 2025 16:42:37 +0000
Subject: [PATCH] disable libunwind on forked children (#19374)

libunwind does not work well after fork, so we disable stack traces on forked processes
---
 src/libnetdata/log/nd_log-init.c              |  2 ++
 src/libnetdata/log/nd_log-internals.h         |  2 ++
 src/libnetdata/log/nd_log-libunwind.c         | 30 +++++++++++++------
 .../spawn_server/spawn_server_nofork.c        |  2 ++
 4 files changed, 27 insertions(+), 9 deletions(-)

diff --git a/src/libnetdata/log/nd_log-init.c b/src/libnetdata/log/nd_log-init.c
index 7f846b136386a3..c3b9523e15cd90 100644
--- a/src/libnetdata/log/nd_log-init.c
+++ b/src/libnetdata/log/nd_log-init.c
@@ -274,6 +274,8 @@ int nd_log_systemd_journal_fd(void) {
 }
 
 void nd_log_reopen_log_files_for_spawn_server(const char *name) {
+    nd_log_forked = true;
+
     gettid_uncached();
 
     if(nd_log.syslog.initialized) {
diff --git a/src/libnetdata/log/nd_log-internals.h b/src/libnetdata/log/nd_log-internals.h
index aa27868a0fd7fe..6105f286add2e6 100644
--- a/src/libnetdata/log/nd_log-internals.h
+++ b/src/libnetdata/log/nd_log-internals.h
@@ -196,6 +196,8 @@ struct log_field;
 const char *errno_annotator(struct log_field *lf);
 const char *priority_annotator(struct log_field *lf);
 const char *timestamp_usec_annotator(struct log_field *lf);
+
+extern bool nd_log_forked;
 bool stack_trace_formatter(BUFFER *wb, void *data);
 
 #if defined(OS_WINDOWS)
diff --git a/src/libnetdata/log/nd_log-libunwind.c b/src/libnetdata/log/nd_log-libunwind.c
index 7ede5a5d199f53..fab1fb9c1c96d5 100644
--- a/src/libnetdata/log/nd_log-libunwind.c
+++ b/src/libnetdata/log/nd_log-libunwind.c
@@ -2,15 +2,25 @@
 
 #include "nd_log-internals.h"
 
+bool nd_log_forked = false;
+
 #ifdef HAVE_LIBUNWIND
 #include <libunwind.h>
 
 bool stack_trace_formatter(BUFFER *wb, void *data __maybe_unused) {
     static __thread bool in_stack_trace = false;
 
-    // Prevent recursion
-    if(in_stack_trace)
-        return buffer_strcat(wb, "stack trace recursion detected"), true;
+    if (nd_log_forked) {
+        // libunwind freezes in forked children
+        buffer_strcat(wb, "stack trace after fork is disabled");
+        return true;
+    }
+
+    if (in_stack_trace) {
+        // Prevent recursion
+        buffer_strcat(wb, "stack trace recursion detected");
+        return true;
+    }
 
     in_stack_trace = true;
 
@@ -23,9 +33,10 @@ bool stack_trace_formatter(BUFFER *wb, void *data __maybe_unused) {
     unw_init_local(&cursor, &context);
 
     // Skip first 3 frames (our logging infrastructure)
-    unw_step(&cursor);
-    unw_step(&cursor);
-    unw_step(&cursor);
+    for (int i = 0; i < 3; i++) {
+        if (unw_step(&cursor) <= 0)
+            goto cleanup; // Ensure proper cleanup if unwinding fails early
+    }
 
     while (unw_step(&cursor) > 0) {
         unw_word_t offset, pc;
@@ -37,17 +48,18 @@ bool stack_trace_formatter(BUFFER *wb, void *data __maybe_unused) {
 
         const char *name = sym;
         if (unw_get_proc_name(&cursor, sym, sizeof(sym), &offset) == 0) {
-            if(frames++) buffer_strcat(wb, "\n");
+            if (frames++) buffer_strcat(wb, "\n");
             buffer_sprintf(wb, "%s+0x%lx", name, (unsigned long)offset);
         }
         else {
-            if(frames++)
+            if (frames++)
                 buffer_strcat(wb, "\n");
             buffer_strcat(wb, "<unknown>");
         }
     }
 
-    in_stack_trace = false;
+cleanup:
+    in_stack_trace = false; // Ensure the flag is reset
     return true;
 }
 
diff --git a/src/libnetdata/spawn_server/spawn_server_nofork.c b/src/libnetdata/spawn_server/spawn_server_nofork.c
index 3fd890f045f2a2..3274de3f531d16 100644
--- a/src/libnetdata/spawn_server/spawn_server_nofork.c
+++ b/src/libnetdata/spawn_server/spawn_server_nofork.c
@@ -1210,6 +1210,8 @@ int spawn_server_exec_kill(SPAWN_SERVER *server, SPAWN_INSTANCE *instance, int t
 }
 
 SPAWN_INSTANCE* spawn_server_exec(SPAWN_SERVER *server, int stderr_fd, int custom_fd, const char **argv, const void *data, size_t data_size, SPAWN_INSTANCE_TYPE type) {
+    if(!server) return NULL;
+
     int pipe_stdin[2] = { -1, -1 }, pipe_stdout[2] = { -1, -1 };
 
     SPAWN_INSTANCE *instance = callocz(1, sizeof(SPAWN_INSTANCE));