RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-05-28 17:35:08 +00:00
Code Issues Activity Actions

Shell: Wait for *any* child to change state when receiving a SIGCHLD

Browse source 
This really just works around the core issue, which is that we have no
reliable way to know exactly who raised the signal (yet).
Fixes #3645, in a very weird (yet apparently standard) way.
This commit is contained in:
AnotherTest 2020-09-30 14:05:12 +03:30 • committed by Andreas Kling
parent a7828434c0
commit 519d1811fd
4 changed files with 68 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

100
Shell/Shell.cpp
View file

@ -60,43 +60,10 @@ extern char** environ;
void Shell::setup_signals()
{
Core::EventLoop::register_signal(SIGCHLD, [this](int) {
Vector<u64> disowned_jobs;
for (auto& it : jobs) {
auto job_id = it.key;
auto& job = *it.value;
int wstatus = 0;
auto child_pid = waitpid(job.pid(), &wstatus, WNOHANG | WUNTRACED);
if (child_pid < 0) {
if (errno == ECHILD) {
// The child process went away before we could process its death, just assume it exited all ok.
// FIXME: This should never happen, the child should stay around until we do the waitpid above.
dbg() << "Child process gone, cannot get exit code for " << job_id;
child_pid = job.pid();
} else {
ASSERT_NOT_REACHED();
}
}
#ifndef __serenity__
if (child_pid == 0) {
// Linux: if child didn't "change state", but existed.
continue;
}
#ifdef SH_DEBUG
dbg() << "SIGCHLD!";
#endif
if (child_pid == job.pid()) {
if (WIFSIGNALED(wstatus) && !WIFSTOPPED(wstatus)) {
job.set_signalled(WTERMSIG(wstatus));
} else if (WIFEXITED(wstatus)) {
job.set_has_exit(WEXITSTATUS(wstatus));
} else if (WIFSTOPPED(wstatus)) {
job.unblock();
job.set_is_suspended(true);
}
}
if (job.should_be_disowned())
disowned_jobs.append(job_id);
}
for (auto job_id : disowned_jobs)
jobs.remove(job_id);
notify_child_event();
});
Core::EventLoop::register_signal(SIGTSTP, [this](auto) {
@ -1413,6 +1380,67 @@ void Shell::custom_event(Core::CustomEvent& event)
}
}
void Shell::notify_child_event()
{
Vector<u64> disowned_jobs;
// Workaround the fact that we can't receive *who* exactly changed state.
// The child might still be alive (and even running) when this signal is dispatched to us
// so just...repeat until we find a suitable child.
// This, of course, will mean that someone can send us a SIGCHILD and we'd be spinning here
// until the next child event we can actually handle.
bool found_child = false;
do {
// Ignore stray SIGCHLD when there are no jobs.
if (jobs.is_empty())
return;
for (auto& it : jobs) {
auto job_id = it.key;
auto& job = *it.value;
int wstatus = 0;
#ifdef SH_DEBUG
dbgf("waitpid({}) = ...", job.pid());
#endif
auto child_pid = waitpid(job.pid(), &wstatus, WNOHANG | WUNTRACED);
#ifdef SH_DEBUG
dbgf("... = {} - {}", child_pid, wstatus);
#endif
if (child_pid < 0) {
if (errno == ECHILD) {
// The child process went away before we could process its death, just assume it exited all ok.
// FIXME: This should never happen, the child should stay around until we do the waitpid above.
dbg() << "Child process gone, cannot get exit code for " << job_id;
child_pid = job.pid();
} else {
ASSERT_NOT_REACHED();
}
}
if (child_pid == 0) {
// If the child existed, but wasn't dead.
continue;
}
if (child_pid == job.pid()) {
if (WIFSIGNALED(wstatus) && !WIFSTOPPED(wstatus)) {
job.set_signalled(WTERMSIG(wstatus));
} else if (WIFEXITED(wstatus)) {
job.set_has_exit(WEXITSTATUS(wstatus));
} else if (WIFSTOPPED(wstatus)) {
job.unblock();
job.set_is_suspended(true);
}
found_child = true;
}
if (job.should_be_disowned())
disowned_jobs.append(job_id);
}
for (auto job_id : disowned_jobs) {
jobs.remove(job_id);
}
} while (!found_child);
}
Shell::Shell(Line::Editor& editor)
: m_editor(editor)
{