blob: 64e1c79745efeebcf6e5606257d76a8e7f26d8bb [file] [log] [blame]
//===-- ---------------------------===//
// The LLVM Compiler Infrastructure
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
// See sanitizer_stoptheworld.h for details.
// This implementation was inspired by Markus Gutschke's
#include "sanitizer_platform.h"
#if SANITIZER_LINUX && (defined(__x86_64__) || defined(__mips__))
#include "sanitizer_stoptheworld.h"
#include "sanitizer_platform_limits_posix.h"
#include "sanitizer_atomic.h"
#include <errno.h>
#include <sched.h> // for CLONE_* definitions
#include <stddef.h>
#include <sys/prctl.h> // for PR_* definitions
#include <sys/ptrace.h> // for PTRACE_* definitions
#include <sys/types.h> // for pid_t
#if SANITIZER_ANDROID && defined(__arm__)
# include <linux/user.h> // for pt_regs
# include <sys/user.h> // for user_regs_struct
#include <sys/wait.h> // for signal-related stuff
#ifdef sa_handler
# undef sa_handler
#ifdef sa_sigaction
# undef sa_sigaction
#include "sanitizer_common.h"
#include "sanitizer_flags.h"
#include "sanitizer_libc.h"
#include "sanitizer_linux.h"
#include "sanitizer_mutex.h"
#include "sanitizer_placement_new.h"
// This module works by spawning a Linux task which then attaches to every
// thread in the caller process with ptrace. This suspends the threads, and
// PTRACE_GETREGS can then be used to obtain their register state. The callback
// supplied to StopTheWorld() is run in the tracer task while the threads are
// suspended.
// The tracer task must be placed in a different thread group for ptrace to
// work, so it cannot be spawned as a pthread. Instead, we use the low-level
// clone() interface (we want to share the address space with the caller
// process, so we prefer clone() over fork()).
// We don't use any libc functions, relying instead on direct syscalls. There
// are two reasons for this:
// 1. calling a library function while threads are suspended could cause a
// deadlock, if one of the treads happens to be holding a libc lock;
// 2. it's generally not safe to call libc functions from the tracer task,
// because clone() does not set up a thread-local storage for it. Any
// thread-local variables used by libc will be shared between the tracer task
// and the thread which spawned it.
COMPILER_CHECK(sizeof(SuspendedThreadID) == sizeof(pid_t));
namespace __sanitizer {
// Structure for passing arguments into the tracer thread.
struct TracerThreadArgument {
StopTheWorldCallback callback;
void *callback_argument;
// The tracer thread waits on this mutex while the parent finishes its
// preparations.
BlockingMutex mutex;
// Tracer thread signals its completion by setting done.
atomic_uintptr_t done;
uptr parent_pid;
// This class handles thread suspending/unsuspending in the tracer thread.
class ThreadSuspender {
explicit ThreadSuspender(pid_t pid, TracerThreadArgument *arg)
: arg(arg)
, pid_(pid) {
CHECK_GE(pid, 0);
bool SuspendAllThreads();
void ResumeAllThreads();
void KillAllThreads();
SuspendedThreadsList &suspended_threads_list() {
return suspended_threads_list_;
TracerThreadArgument *arg;
SuspendedThreadsList suspended_threads_list_;
pid_t pid_;
bool SuspendThread(SuspendedThreadID thread_id);
bool ThreadSuspender::SuspendThread(SuspendedThreadID tid) {
// Are we already attached to this thread?
// Currently this check takes linear time, however the number of threads is
// usually small.
if (suspended_threads_list_.Contains(tid))
return false;
int pterrno;
if (internal_iserror(internal_ptrace(PTRACE_ATTACH, tid, NULL, NULL),
&pterrno)) {
// Either the thread is dead, or something prevented us from attaching.
// Log this event and move on.
VReport(1, "Could not attach to thread %d (errno %d).\n", tid, pterrno);
return false;
} else {
VReport(2, "Attached to thread %d.\n", tid);
// The thread is not guaranteed to stop before ptrace returns, so we must
// wait on it. Note: if the thread receives a signal concurrently,
// we can get notification about the signal before notification about stop.
// In such case we need to forward the signal to the thread, otherwise
// the signal will be missed (as we do PTRACE_DETACH with arg=0) and
// any logic relying on signals will break. After forwarding we need to
// continue to wait for stopping, because the thread is not stopped yet.
// We do ignore delivery of SIGSTOP, because we want to make stop-the-world
// as invisible as possible.
for (;;) {
int status;
uptr waitpid_status;
HANDLE_EINTR(waitpid_status, internal_waitpid(tid, &status, __WALL));
int wperrno;
if (internal_iserror(waitpid_status, &wperrno)) {
// Got a ECHILD error. I don't think this situation is possible, but it
// doesn't hurt to report it.
VReport(1, "Waiting on thread %d failed, detaching (errno %d).\n",
tid, wperrno);
internal_ptrace(PTRACE_DETACH, tid, NULL, NULL);
return false;
if (WIFSTOPPED(status) && WSTOPSIG(status) != SIGSTOP) {
internal_ptrace(PTRACE_CONT, tid, 0, (void*)(uptr)WSTOPSIG(status));
return true;
void ThreadSuspender::ResumeAllThreads() {
for (uptr i = 0; i < suspended_threads_list_.thread_count(); i++) {
pid_t tid = suspended_threads_list_.GetThreadID(i);
int pterrno;
if (!internal_iserror(internal_ptrace(PTRACE_DETACH, tid, NULL, NULL),
&pterrno)) {
VReport(2, "Detached from thread %d.\n", tid);
} else {
// Either the thread is dead, or we are already detached.
// The latter case is possible, for instance, if this function was called
// from a signal handler.
VReport(1, "Could not detach from thread %d (errno %d).\n", tid, pterrno);
void ThreadSuspender::KillAllThreads() {
for (uptr i = 0; i < suspended_threads_list_.thread_count(); i++)
internal_ptrace(PTRACE_KILL, suspended_threads_list_.GetThreadID(i),
bool ThreadSuspender::SuspendAllThreads() {
ThreadLister thread_lister(pid_);
bool added_threads;
do {
// Run through the directory entries once.
added_threads = false;
pid_t tid = thread_lister.GetNextTID();
while (tid >= 0) {
if (SuspendThread(tid))
added_threads = true;
tid = thread_lister.GetNextTID();
if (thread_lister.error()) {
// Detach threads and fail.
return false;
} while (added_threads);
return true;
// Pointer to the ThreadSuspender instance for use in signal handler.
static ThreadSuspender *thread_suspender_instance = NULL;
// Synchronous signals that should not be blocked.
static const int kSyncSignals[] = { SIGABRT, SIGILL, SIGFPE, SIGSEGV, SIGBUS,
static DieCallbackType old_die_callback;
// Signal handler to wake up suspended threads when the tracer thread dies.
static void TracerThreadSignalHandler(int signum, void *siginfo, void *uctx) {
SignalContext ctx = SignalContext::Create(siginfo, uctx);
VPrintf(1, "Tracer caught signal %d: addr=0x%zx pc=0x%zx sp=0x%zx\n",
signum, ctx.addr, ctx.pc, ctx.sp);
ThreadSuspender *inst = thread_suspender_instance;
if (inst != NULL) {
if (signum == SIGABRT)
old_die_callback = NULL;
thread_suspender_instance = NULL;
atomic_store(&inst->arg->done, 1, memory_order_relaxed);
internal__exit((signum == SIGABRT) ? 1 : 2);
static void TracerThreadDieCallback() {
// Generally a call to Die() in the tracer thread should be fatal to the
// parent process as well, because they share the address space.
// This really only works correctly if all the threads are suspended at this
// point. So we correctly handle calls to Die() from within the callback, but
// not those that happen before or after the callback. Hopefully there aren't
// a lot of opportunities for that to happen...
ThreadSuspender *inst = thread_suspender_instance;
if (inst != NULL && stoptheworld_tracer_pid == internal_getpid()) {
thread_suspender_instance = NULL;
if (old_die_callback)
old_die_callback = NULL;
// Size of alternative stack for signal handlers in the tracer thread.
static const int kHandlerStackSize = 4096;
// This function will be run as a cloned task.
static int TracerThread(void* argument) {
TracerThreadArgument *tracer_thread_argument =
(TracerThreadArgument *)argument;
internal_prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0);
// Check if parent is already dead.
if (internal_getppid() != tracer_thread_argument->parent_pid)
// Wait for the parent thread to finish preparations.
old_die_callback = GetDieCallback();
ThreadSuspender thread_suspender(internal_getppid(), tracer_thread_argument);
// Global pointer for the signal handler.
thread_suspender_instance = &thread_suspender;
// Alternate stack for signal handling.
InternalScopedBuffer<char> handler_stack_memory(kHandlerStackSize);
struct sigaltstack handler_stack;
internal_memset(&handler_stack, 0, sizeof(handler_stack));
handler_stack.ss_sp =;
handler_stack.ss_size = kHandlerStackSize;
internal_sigaltstack(&handler_stack, NULL);
// Install our handler for synchronous signals. Other signals should be
// blocked by the mask we inherited from the parent thread.
for (uptr i = 0; i < ARRAY_SIZE(kSyncSignals); i++) {
__sanitizer_sigaction act;
internal_memset(&act, 0, sizeof(act));
act.sigaction = TracerThreadSignalHandler;
act.sa_flags = SA_ONSTACK | SA_SIGINFO;
internal_sigaction_norestorer(kSyncSignals[i], &act, 0);
int exit_code = 0;
if (!thread_suspender.SuspendAllThreads()) {
VReport(1, "Failed suspending threads.\n");
exit_code = 3;
} else {
exit_code = 0;
thread_suspender_instance = NULL;
atomic_store(&tracer_thread_argument->done, 1, memory_order_relaxed);
return exit_code;
class ScopedStackSpaceWithGuard {
explicit ScopedStackSpaceWithGuard(uptr stack_size) {
stack_size_ = stack_size;
guard_size_ = GetPageSizeCached();
// FIXME: Omitting MAP_STACK here works in current kernels but might break
// in the future.
guard_start_ = (uptr)MmapOrDie(stack_size_ + guard_size_,
CHECK_EQ(guard_start_, (uptr)Mprotect((uptr)guard_start_, guard_size_));
~ScopedStackSpaceWithGuard() {
UnmapOrDie((void *)guard_start_, stack_size_ + guard_size_);
void *Bottom() const {
return (void *)(guard_start_ + stack_size_ + guard_size_);
uptr stack_size_;
uptr guard_size_;
uptr guard_start_;
// We have a limitation on the stack frame size, so some stuff had to be moved
// into globals.
static __sanitizer_sigset_t blocked_sigset;
static __sanitizer_sigset_t old_sigset;
class StopTheWorldScope {
StopTheWorldScope() {
// Make this process dumpable. Processes that are not dumpable cannot be
// attached to.
process_was_dumpable_ = internal_prctl(PR_GET_DUMPABLE, 0, 0, 0, 0);
if (!process_was_dumpable_)
internal_prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
~StopTheWorldScope() {
// Restore the dumpable flag.
if (!process_was_dumpable_)
internal_prctl(PR_SET_DUMPABLE, 0, 0, 0, 0);
int process_was_dumpable_;
// When sanitizer output is being redirected to file (i.e. by using log_path),
// the tracer should write to the parent's log instead of trying to open a new
// file. Alert the logging code to the fact that we have a tracer.
struct ScopedSetTracerPID {
explicit ScopedSetTracerPID(uptr tracer_pid) {
stoptheworld_tracer_pid = tracer_pid;
stoptheworld_tracer_ppid = internal_getpid();
~ScopedSetTracerPID() {
stoptheworld_tracer_pid = 0;
stoptheworld_tracer_ppid = 0;
void StopTheWorld(StopTheWorldCallback callback, void *argument) {
StopTheWorldScope in_stoptheworld;
// Prepare the arguments for TracerThread.
struct TracerThreadArgument tracer_thread_argument;
tracer_thread_argument.callback = callback;
tracer_thread_argument.callback_argument = argument;
tracer_thread_argument.parent_pid = internal_getpid();
atomic_store(&tracer_thread_argument.done, 0, memory_order_relaxed);
const uptr kTracerStackSize = 2 * 1024 * 1024;
ScopedStackSpaceWithGuard tracer_stack(kTracerStackSize);
// Block the execution of TracerThread until after we have set ptrace
// permissions.
// Signal handling story.
// We don't want async signals to be delivered to the tracer thread,
// so we block all async signals before creating the thread. An async signal
// handler can temporary modify errno, which is shared with this thread.
// We ought to use pthread_sigmask here, because sigprocmask has undefined
// behavior in multithreaded programs. However, on linux sigprocmask is
// equivalent to pthread_sigmask with the exception that pthread_sigmask
// does not allow to block some signals used internally in pthread
// implementation. We are fine with blocking them here, we are really not
// going to pthread_cancel the thread.
// The tracer thread should not raise any synchronous signals. But in case it
// does, we setup a special handler for sync signals that properly kills the
// parent as well. Note: we don't pass CLONE_SIGHAND to clone, so handlers
// in the tracer thread won't interfere with user program. Double note: if a
// user does something along the lines of 'kill -11 pid', that can kill the
// process even if user setup own handler for SEGV.
// Thing to watch out for: this code should not change behavior of user code
// in any observable way. In particular it should not override user signal
// handlers.
for (uptr i = 0; i < ARRAY_SIZE(kSyncSignals); i++)
internal_sigdelset(&blocked_sigset, kSyncSignals[i]);
int rv = internal_sigprocmask(SIG_BLOCK, &blocked_sigset, &old_sigset);
CHECK_EQ(rv, 0);
uptr tracer_pid = internal_clone(
TracerThread, tracer_stack.Bottom(),
&tracer_thread_argument, 0 /* parent_tidptr */, 0 /* newtls */, 0
/* child_tidptr */);
internal_sigprocmask(SIG_SETMASK, &old_sigset, 0);
int local_errno = 0;
if (internal_iserror(tracer_pid, &local_errno)) {
VReport(1, "Failed spawning a tracer thread (errno %d).\n", local_errno);
} else {
ScopedSetTracerPID scoped_set_tracer_pid(tracer_pid);
// On some systems we have to explicitly declare that we want to be traced
// by the tracer thread.
internal_prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
// Allow the tracer thread to start.
// NOTE: errno is shared between this thread and the tracer thread.
// internal_waitpid() may call syscall() which can access/spoil errno,
// so we can't call it now. Instead we for the tracer thread to finish using
// the spin loop below. Man page for sched_yield() says "In the Linux
// implementation, sched_yield() always succeeds", so let's hope it does not
// spoil errno. Note that this spin loop runs only for brief periods before
// the tracer thread has suspended us and when it starts unblocking threads.
while (atomic_load(&tracer_thread_argument.done, memory_order_relaxed) == 0)
// Now the tracer thread is about to exit and does not touch errno,
// wait for it.
for (;;) {
uptr waitpid_status = internal_waitpid(tracer_pid, NULL, __WALL);
if (!internal_iserror(waitpid_status, &local_errno))
if (local_errno == EINTR)
VReport(1, "Waiting on the tracer thread failed (errno %d).\n",
// Platform-specific methods from SuspendedThreadsList.
#if SANITIZER_ANDROID && defined(__arm__)
typedef pt_regs regs_struct;
#define REG_SP ARM_sp
#elif SANITIZER_LINUX && defined(__arm__)
typedef user_regs regs_struct;
#define REG_SP uregs[13]
#elif defined(__i386__) || defined(__x86_64__)
typedef user_regs_struct regs_struct;
#if defined(__i386__)
#define REG_SP esp
#define REG_SP rsp
#elif defined(__powerpc__) || defined(__powerpc64__)
typedef pt_regs regs_struct;
#define REG_SP gpr[PT_R1]
#elif defined(__mips__)
typedef struct user regs_struct;
#define REG_SP regs[EF_REG29]
#error "Unsupported architecture"
#endif // SANITIZER_ANDROID && defined(__arm__)
int SuspendedThreadsList::GetRegistersAndSP(uptr index,
uptr *buffer,
uptr *sp) const {
pid_t tid = GetThreadID(index);
regs_struct regs;
int pterrno;
if (internal_iserror(internal_ptrace(PTRACE_GETREGS, tid, NULL, &regs),
&pterrno)) {
VReport(1, "Could not get registers from thread %d (errno %d).\n", tid,
return -1;
*sp = regs.REG_SP;
internal_memcpy(buffer, &regs, sizeof(regs));
return 0;
uptr SuspendedThreadsList::RegisterCount() {
return sizeof(regs_struct) / sizeof(uptr);
} // namespace __sanitizer
#endif // SANITIZER_LINUX && (defined(__x86_64__) || defined(__mips__))