android: restore the default SIGABRT handler in fatal_error()

async_safe_fatal() calls abort() at the end, which can be intercepted by a custom SIGABRT handler.

In particular, crashlytics installs such a handler and tries to fork() after catching SIGABRT.

hardened_malloc uses pthread_atfork() to register fork handlers. These handlers try to lock internal
hardened_malloc mutexes. If at least one of those mutexes is already locked, which is usually the
case, thread that called fatai_error() gets deadlocked, while the other threads (if there are any)
continue to run.
pull/238/head 2023123100-redfin
Dmitry Muhomor 2023-12-31 15:37:32 +02:00 committed by Daniel Micay
parent 7093fdc482
commit 365ee6900d
1 changed files with 3 additions and 0 deletions

3
util.c
View File

@ -6,6 +6,8 @@
#ifdef __ANDROID__
#include <async_safe/log.h>
int mallopt(int param, int value);
#define M_BIONIC_RESTORE_DEFAULT_SIGABRT_HANDLER (-1003)
#endif
#include "util.h"
@ -30,6 +32,7 @@ static int write_full(int fd, const char *buf, size_t length) {
COLD noreturn void fatal_error(const char *s) {
#ifdef __ANDROID__
mallopt(M_BIONIC_RESTORE_DEFAULT_SIGABRT_HANDLER, 0);
async_safe_fatal("hardened_malloc: fatal allocator error: %s", s);
#else
const char *prefix = "fatal allocator error: ";