disable libunwind on forked children (#19374)
libunwind does not work well after fork, so we disable stack traces on forked processes
Costa Tsaousis committed
Jan 10, 2025 at 16:42 UTC
8e127e75527a02c2b3854f94b419cf8138a378cf
4 files changed
+27
-9
src/libnetdata/log/nd_log-init.c
+2
@@ -274,6 +274,8 @@ int nd_log_systemd_journal_fd(void) {
274
}
275
276
void nd_log_reopen_log_files_for_spawn_server(const char *name) {
277
+ nd_log_forked = true;
278
+
279
gettid_uncached();
280
281
if(nd_log.syslog.initialized) {
src/libnetdata/log/nd_log-internals.h
+2
@@ -196,6 +196,8 @@ struct log_field;
196
const char *errno_annotator(struct log_field *lf);
197
const char *priority_annotator(struct log_field *lf);
198
const char *timestamp_usec_annotator(struct log_field *lf);
199
+
200
+extern bool nd_log_forked;
201
bool stack_trace_formatter(BUFFER *wb, void *data);
202
203
#if defined(OS_WINDOWS)
src/libnetdata/log/nd_log-libunwind.c
+21
-9
@@ -2,15 +2,25 @@
2
3
#include "nd_log-internals.h"
4
5
+bool nd_log_forked = false;
6
+
7
#ifdef HAVE_LIBUNWIND
8
#include <libunwind.h>
9
10
bool stack_trace_formatter(BUFFER *wb, void *data __maybe_unused) {
11
static __thread bool in_stack_trace = false;
12
11
- // Prevent recursion
12
- if(in_stack_trace)
13
- return buffer_strcat(wb, "stack trace recursion detected"), true;
13
+ if (nd_log_forked) {
14
+ // libunwind freezes in forked children
15
+ buffer_strcat(wb, "stack trace after fork is disabled");
16
+ return true;
17
+ }
18
+
19
+ if (in_stack_trace) {
20
+ // Prevent recursion
21
+ buffer_strcat(wb, "stack trace recursion detected");
22
+ return true;
23
+ }
24
25
in_stack_trace = true;
26
@@ -23,9 +33,10 @@ bool stack_trace_formatter(BUFFER *wb, void *data __maybe_unused) {
33
unw_init_local(&cursor, &context);
34
35
// Skip first 3 frames (our logging infrastructure)
26
- unw_step(&cursor);
27
- unw_step(&cursor);
28
- unw_step(&cursor);
36
+ for (int i = 0; i < 3; i++) {
37
+ if (unw_step(&cursor) <= 0)
38
+ goto cleanup; // Ensure proper cleanup if unwinding fails early
39
+ }
40
41
while (unw_step(&cursor) > 0) {
42
unw_word_t offset, pc;
@@ -37,17 +48,18 @@ bool stack_trace_formatter(BUFFER *wb, void *data __maybe_unused) {
48
49
const char *name = sym;
50
if (unw_get_proc_name(&cursor, sym, sizeof(sym), &offset) == 0) {
40
- if(frames++) buffer_strcat(wb, "\n");
51
+ if (frames++) buffer_strcat(wb, "\n");
52
buffer_sprintf(wb, "%s+0x%lx", name, (unsigned long)offset);
53
}
54
else {
44
- if(frames++)
55
+ if (frames++)
56
buffer_strcat(wb, "\n");
57
buffer_strcat(wb, "<unknown>");
58
}
59
}
60
50
- in_stack_trace = false;
61
+cleanup:
62
+ in_stack_trace = false; // Ensure the flag is reset
63
return true;
64
}
65
src/libnetdata/spawn_server/spawn_server_nofork.c
+2
@@ -1210,6 +1210,8 @@ int spawn_server_exec_kill(SPAWN_SERVER *server, SPAWN_INSTANCE *instance, int t
1210
}
1211
1212
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) {
1213
+ if(!server) return NULL;
1214
+
1215
int pipe_stdin[2] = { -1, -1 }, pipe_stdout[2] = { -1, -1 };
1216
1217
SPAWN_INSTANCE *instance = callocz(1, sizeof(SPAWN_INSTANCE));