1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/tracehook.h>
53 #include <linux/fs_struct.h>
54 #include <linux/init_task.h>
55 #include <linux/perf_event.h>
56 #include <trace/events/sched.h>
57 #include <linux/hw_breakpoint.h>
58 #include <linux/oom.h>
59 #include <linux/writeback.h>
60 #include <linux/shm.h>
61 #include <linux/kcov.h>
62 #include <linux/random.h>
63 #include <linux/rcuwait.h>
64 #include <linux/compat.h>
65 #include <linux/io_uring.h>
67 #include <linux/uaccess.h>
68 #include <asm/unistd.h>
69 #include <asm/mmu_context.h>
71 static void __unhash_process(struct task_struct *p, bool group_dead)
74 detach_pid(p, PIDTYPE_PID);
76 detach_pid(p, PIDTYPE_TGID);
77 detach_pid(p, PIDTYPE_PGID);
78 detach_pid(p, PIDTYPE_SID);
80 list_del_rcu(&p->tasks);
81 list_del_init(&p->sibling);
82 __this_cpu_dec(process_counts);
84 list_del_rcu(&p->thread_group);
85 list_del_rcu(&p->thread_node);
89 * This function expects the tasklist_lock write-locked.
91 static void __exit_signal(struct task_struct *tsk)
93 struct signal_struct *sig = tsk->signal;
94 bool group_dead = thread_group_leader(tsk);
95 struct sighand_struct *sighand;
96 struct tty_struct *tty;
99 sighand = rcu_dereference_check(tsk->sighand,
100 lockdep_tasklist_lock_is_held());
101 spin_lock(&sighand->siglock);
103 #ifdef CONFIG_POSIX_TIMERS
104 posix_cpu_timers_exit(tsk);
106 posix_cpu_timers_exit_group(tsk);
114 * If there is any task waiting for the group exit
117 if (sig->notify_count > 0 && !--sig->notify_count)
118 wake_up_process(sig->group_exit_task);
120 if (tsk == sig->curr_target)
121 sig->curr_target = next_thread(tsk);
124 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
125 sizeof(unsigned long long));
128 * Accumulate here the counters for all threads as they die. We could
129 * skip the group leader because it is the last user of signal_struct,
130 * but we want to avoid the race with thread_group_cputime() which can
131 * see the empty ->thread_head list.
133 task_cputime(tsk, &utime, &stime);
134 write_seqlock(&sig->stats_lock);
137 sig->gtime += task_gtime(tsk);
138 sig->min_flt += tsk->min_flt;
139 sig->maj_flt += tsk->maj_flt;
140 sig->nvcsw += tsk->nvcsw;
141 sig->nivcsw += tsk->nivcsw;
142 sig->inblock += task_io_get_inblock(tsk);
143 sig->oublock += task_io_get_oublock(tsk);
144 task_io_accounting_add(&sig->ioac, &tsk->ioac);
145 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
147 __unhash_process(tsk, group_dead);
148 write_sequnlock(&sig->stats_lock);
151 * Do this under ->siglock, we can race with another thread
152 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
154 flush_sigqueue(&tsk->pending);
156 spin_unlock(&sighand->siglock);
158 __cleanup_sighand(sighand);
159 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
161 flush_sigqueue(&sig->shared_pending);
166 static void delayed_put_task_struct(struct rcu_head *rhp)
168 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
170 perf_event_delayed_put(tsk);
171 trace_sched_process_free(tsk);
172 put_task_struct(tsk);
175 void put_task_struct_rcu_user(struct task_struct *task)
177 if (refcount_dec_and_test(&task->rcu_users))
178 call_rcu(&task->rcu, delayed_put_task_struct);
181 void release_task(struct task_struct *p)
183 struct task_struct *leader;
184 struct pid *thread_pid;
187 /* don't need to get the RCU readlock here - the process is dead and
188 * can't be modifying its own credentials. But shut RCU-lockdep up */
190 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
195 write_lock_irq(&tasklist_lock);
196 ptrace_release_task(p);
197 thread_pid = get_pid(p->thread_pid);
201 * If we are the last non-leader member of the thread
202 * group, and the leader is zombie, then notify the
203 * group leader's parent process. (if it wants notification.)
206 leader = p->group_leader;
207 if (leader != p && thread_group_empty(leader)
208 && leader->exit_state == EXIT_ZOMBIE) {
210 * If we were the last child thread and the leader has
211 * exited already, and the leader's parent ignores SIGCHLD,
212 * then we are the one who should release the leader.
214 zap_leader = do_notify_parent(leader, leader->exit_signal);
216 leader->exit_state = EXIT_DEAD;
219 write_unlock_irq(&tasklist_lock);
220 seccomp_filter_release(p);
221 proc_flush_pid(thread_pid);
224 put_task_struct_rcu_user(p);
227 if (unlikely(zap_leader))
231 int rcuwait_wake_up(struct rcuwait *w)
234 struct task_struct *task;
239 * Order condition vs @task, such that everything prior to the load
240 * of @task is visible. This is the condition as to why the user called
241 * rcuwait_wake() in the first place. Pairs with set_current_state()
242 * barrier (A) in rcuwait_wait_event().
245 * [S] tsk = current [S] cond = true
251 task = rcu_dereference(w->task);
253 ret = wake_up_process(task);
258 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
261 * Determine if a process group is "orphaned", according to the POSIX
262 * definition in 2.2.2.52. Orphaned process groups are not to be affected
263 * by terminal-generated stop signals. Newly orphaned process groups are
264 * to receive a SIGHUP and a SIGCONT.
266 * "I ask you, have you ever known what it is to be an orphan?"
268 static int will_become_orphaned_pgrp(struct pid *pgrp,
269 struct task_struct *ignored_task)
271 struct task_struct *p;
273 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
274 if ((p == ignored_task) ||
275 (p->exit_state && thread_group_empty(p)) ||
276 is_global_init(p->real_parent))
279 if (task_pgrp(p->real_parent) != pgrp &&
280 task_session(p->real_parent) == task_session(p))
282 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
287 int is_current_pgrp_orphaned(void)
291 read_lock(&tasklist_lock);
292 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
293 read_unlock(&tasklist_lock);
298 static bool has_stopped_jobs(struct pid *pgrp)
300 struct task_struct *p;
302 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303 if (p->signal->flags & SIGNAL_STOP_STOPPED)
305 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
311 * Check to see if any process groups have become orphaned as
312 * a result of our exiting, and if they have any stopped jobs,
313 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
316 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
318 struct pid *pgrp = task_pgrp(tsk);
319 struct task_struct *ignored_task = tsk;
322 /* exit: our father is in a different pgrp than
323 * we are and we were the only connection outside.
325 parent = tsk->real_parent;
327 /* reparent: our child is in a different pgrp than
328 * we are, and it was the only connection outside.
332 if (task_pgrp(parent) != pgrp &&
333 task_session(parent) == task_session(tsk) &&
334 will_become_orphaned_pgrp(pgrp, ignored_task) &&
335 has_stopped_jobs(pgrp)) {
336 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
337 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
343 * A task is exiting. If it owned this mm, find a new owner for the mm.
345 void mm_update_next_owner(struct mm_struct *mm)
347 struct task_struct *c, *g, *p = current;
351 * If the exiting or execing task is not the owner, it's
352 * someone else's problem.
357 * The current owner is exiting/execing and there are no other
358 * candidates. Do not leave the mm pointing to a possibly
359 * freed task structure.
361 if (atomic_read(&mm->mm_users) <= 1) {
362 WRITE_ONCE(mm->owner, NULL);
366 read_lock(&tasklist_lock);
368 * Search in the children
370 list_for_each_entry(c, &p->children, sibling) {
372 goto assign_new_owner;
376 * Search in the siblings
378 list_for_each_entry(c, &p->real_parent->children, sibling) {
380 goto assign_new_owner;
384 * Search through everything else, we should not get here often.
386 for_each_process(g) {
387 if (g->flags & PF_KTHREAD)
389 for_each_thread(g, c) {
391 goto assign_new_owner;
396 read_unlock(&tasklist_lock);
398 * We found no owner yet mm_users > 1: this implies that we are
399 * most likely racing with swapoff (try_to_unuse()) or /proc or
400 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
402 WRITE_ONCE(mm->owner, NULL);
409 * The task_lock protects c->mm from changing.
410 * We always want mm->owner->mm == mm
414 * Delay read_unlock() till we have the task_lock()
415 * to ensure that c does not slip away underneath us
417 read_unlock(&tasklist_lock);
423 WRITE_ONCE(mm->owner, c);
427 #endif /* CONFIG_MEMCG */
430 * Turn us into a lazy TLB process if we
433 static void exit_mm(void)
435 struct mm_struct *mm = current->mm;
436 struct core_state *core_state;
438 exit_mm_release(current, mm);
443 * Serialize with any possible pending coredump.
444 * We must hold mmap_lock around checking core_state
445 * and clearing tsk->mm. The core-inducing thread
446 * will increment ->nr_threads for each thread in the
447 * group with ->mm != NULL.
450 core_state = mm->core_state;
452 struct core_thread self;
454 mmap_read_unlock(mm);
457 if (self.task->flags & PF_SIGNALED)
458 self.next = xchg(&core_state->dumper.next, &self);
462 * Implies mb(), the result of xchg() must be visible
463 * to core_state->dumper.
465 if (atomic_dec_and_test(&core_state->nr_threads))
466 complete(&core_state->startup);
469 set_current_state(TASK_UNINTERRUPTIBLE);
470 if (!self.task) /* see coredump_finish() */
472 freezable_schedule();
474 __set_current_state(TASK_RUNNING);
478 BUG_ON(mm != current->active_mm);
479 /* more a memory barrier than a real lock */
482 * When a thread stops operating on an address space, the loop
483 * in membarrier_private_expedited() may not observe that
484 * tsk->mm, and the loop in membarrier_global_expedited() may
485 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
486 * rq->membarrier_state, so those would not issue an IPI.
487 * Membarrier requires a memory barrier after accessing
488 * user-space memory, before clearing tsk->mm or the
489 * rq->membarrier_state.
491 smp_mb__after_spinlock();
494 membarrier_update_current_mm(NULL);
495 enter_lazy_tlb(mm, current);
497 task_unlock(current);
498 mmap_read_unlock(mm);
499 mm_update_next_owner(mm);
501 if (test_thread_flag(TIF_MEMDIE))
505 static struct task_struct *find_alive_thread(struct task_struct *p)
507 struct task_struct *t;
509 for_each_thread(p, t) {
510 if (!(t->flags & PF_EXITING))
516 static struct task_struct *find_child_reaper(struct task_struct *father,
517 struct list_head *dead)
518 __releases(&tasklist_lock)
519 __acquires(&tasklist_lock)
521 struct pid_namespace *pid_ns = task_active_pid_ns(father);
522 struct task_struct *reaper = pid_ns->child_reaper;
523 struct task_struct *p, *n;
525 if (likely(reaper != father))
528 reaper = find_alive_thread(father);
530 pid_ns->child_reaper = reaper;
534 write_unlock_irq(&tasklist_lock);
536 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
537 list_del_init(&p->ptrace_entry);
541 zap_pid_ns_processes(pid_ns);
542 write_lock_irq(&tasklist_lock);
548 * When we die, we re-parent all our children, and try to:
549 * 1. give them to another thread in our thread group, if such a member exists
550 * 2. give it to the first ancestor process which prctl'd itself as a
551 * child_subreaper for its children (like a service manager)
552 * 3. give it to the init process (PID 1) in our pid namespace
554 static struct task_struct *find_new_reaper(struct task_struct *father,
555 struct task_struct *child_reaper)
557 struct task_struct *thread, *reaper;
559 thread = find_alive_thread(father);
563 if (father->signal->has_child_subreaper) {
564 unsigned int ns_level = task_pid(father)->level;
566 * Find the first ->is_child_subreaper ancestor in our pid_ns.
567 * We can't check reaper != child_reaper to ensure we do not
568 * cross the namespaces, the exiting parent could be injected
569 * by setns() + fork().
570 * We check pid->level, this is slightly more efficient than
571 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
573 for (reaper = father->real_parent;
574 task_pid(reaper)->level == ns_level;
575 reaper = reaper->real_parent) {
576 if (reaper == &init_task)
578 if (!reaper->signal->is_child_subreaper)
580 thread = find_alive_thread(reaper);
590 * Any that need to be release_task'd are put on the @dead list.
592 static void reparent_leader(struct task_struct *father, struct task_struct *p,
593 struct list_head *dead)
595 if (unlikely(p->exit_state == EXIT_DEAD))
598 /* We don't want people slaying init. */
599 p->exit_signal = SIGCHLD;
601 /* If it has exited notify the new parent about this child's death. */
603 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
604 if (do_notify_parent(p, p->exit_signal)) {
605 p->exit_state = EXIT_DEAD;
606 list_add(&p->ptrace_entry, dead);
610 kill_orphaned_pgrp(p, father);
614 * This does two things:
616 * A. Make init inherit all the child processes
617 * B. Check to see if any process groups have become orphaned
618 * as a result of our exiting, and if they have any stopped
619 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
621 static void forget_original_parent(struct task_struct *father,
622 struct list_head *dead)
624 struct task_struct *p, *t, *reaper;
626 if (unlikely(!list_empty(&father->ptraced)))
627 exit_ptrace(father, dead);
629 /* Can drop and reacquire tasklist_lock */
630 reaper = find_child_reaper(father, dead);
631 if (list_empty(&father->children))
634 reaper = find_new_reaper(father, reaper);
635 list_for_each_entry(p, &father->children, sibling) {
636 for_each_thread(p, t) {
637 RCU_INIT_POINTER(t->real_parent, reaper);
638 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
639 if (likely(!t->ptrace))
640 t->parent = t->real_parent;
641 if (t->pdeath_signal)
642 group_send_sig_info(t->pdeath_signal,
647 * If this is a threaded reparent there is no need to
648 * notify anyone anything has happened.
650 if (!same_thread_group(reaper, father))
651 reparent_leader(father, p, dead);
653 list_splice_tail_init(&father->children, &reaper->children);
657 * Send signals to all our closest relatives so that they know
658 * to properly mourn us..
660 static void exit_notify(struct task_struct *tsk, int group_dead)
663 struct task_struct *p, *n;
666 write_lock_irq(&tasklist_lock);
667 forget_original_parent(tsk, &dead);
670 kill_orphaned_pgrp(tsk->group_leader, NULL);
672 tsk->exit_state = EXIT_ZOMBIE;
673 if (unlikely(tsk->ptrace)) {
674 int sig = thread_group_leader(tsk) &&
675 thread_group_empty(tsk) &&
676 !ptrace_reparented(tsk) ?
677 tsk->exit_signal : SIGCHLD;
678 autoreap = do_notify_parent(tsk, sig);
679 } else if (thread_group_leader(tsk)) {
680 autoreap = thread_group_empty(tsk) &&
681 do_notify_parent(tsk, tsk->exit_signal);
687 tsk->exit_state = EXIT_DEAD;
688 list_add(&tsk->ptrace_entry, &dead);
691 /* mt-exec, de_thread() is waiting for group leader */
692 if (unlikely(tsk->signal->notify_count < 0))
693 wake_up_process(tsk->signal->group_exit_task);
694 write_unlock_irq(&tasklist_lock);
696 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
697 list_del_init(&p->ptrace_entry);
702 #ifdef CONFIG_DEBUG_STACK_USAGE
703 static void check_stack_usage(void)
705 static DEFINE_SPINLOCK(low_water_lock);
706 static int lowest_to_date = THREAD_SIZE;
709 free = stack_not_used(current);
711 if (free >= lowest_to_date)
714 spin_lock(&low_water_lock);
715 if (free < lowest_to_date) {
716 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
717 current->comm, task_pid_nr(current), free);
718 lowest_to_date = free;
720 spin_unlock(&low_water_lock);
723 static inline void check_stack_usage(void) {}
726 void __noreturn do_exit(long code)
728 struct task_struct *tsk = current;
732 * We can get here from a kernel oops, sometimes with preemption off.
733 * Start by checking for critical errors.
734 * Then fix up important state like USER_DS and preemption.
735 * Then do everything else.
738 WARN_ON(blk_needs_flush_plug(tsk));
740 if (unlikely(in_interrupt()))
741 panic("Aiee, killing interrupt handler!");
742 if (unlikely(!tsk->pid))
743 panic("Attempted to kill the idle task!");
746 * If do_exit is called because this processes oopsed, it's possible
747 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
748 * continuing. Amongst other possible reasons, this is to prevent
749 * mm_release()->clear_child_tid() from writing to a user-controlled
752 force_uaccess_begin();
754 if (unlikely(in_atomic())) {
755 pr_info("note: %s[%d] exited with preempt_count %d\n",
756 current->comm, task_pid_nr(current),
758 preempt_count_set(PREEMPT_ENABLED);
761 profile_task_exit(tsk);
764 ptrace_event(PTRACE_EVENT_EXIT, code);
766 validate_creds_for_do_exit(tsk);
769 * We're taking recursive faults here in do_exit. Safest is to just
770 * leave this task alone and wait for reboot.
772 if (unlikely(tsk->flags & PF_EXITING)) {
773 pr_alert("Fixing recursive fault but reboot is needed!\n");
774 futex_exit_recursive(tsk);
775 set_current_state(TASK_UNINTERRUPTIBLE);
779 io_uring_files_cancel();
780 exit_signals(tsk); /* sets PF_EXITING */
782 /* sync mm's RSS info before statistics gathering */
784 sync_mm_rss(tsk->mm);
785 acct_update_integrals(tsk);
786 group_dead = atomic_dec_and_test(&tsk->signal->live);
789 * If the last thread of global init has exited, panic
790 * immediately to get a useable coredump.
792 if (unlikely(is_global_init(tsk)))
793 panic("Attempted to kill init! exitcode=0x%08x\n",
794 tsk->signal->group_exit_code ?: (int)code);
796 #ifdef CONFIG_POSIX_TIMERS
797 hrtimer_cancel(&tsk->signal->real_timer);
798 exit_itimers(tsk->signal);
801 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
803 acct_collect(code, group_dead);
808 tsk->exit_code = code;
809 taskstats_exit(tsk, group_dead);
815 trace_sched_process_exit(tsk);
822 disassociate_ctty(1);
823 exit_task_namespaces(tsk);
828 * Flush inherited counters to the parent - before the parent
829 * gets woken up by child-exit notifications.
831 * because of cgroup mode, must be called before cgroup_exit()
833 perf_event_exit_task(tsk);
835 sched_autogroup_exit_task(tsk);
839 * FIXME: do that only when needed, using sched_exit tracepoint
841 flush_ptrace_hw_breakpoint(tsk);
843 exit_tasks_rcu_start();
844 exit_notify(tsk, group_dead);
845 proc_exit_connector(tsk);
846 mpol_put_task_policy(tsk);
848 if (unlikely(current->pi_state_cache))
849 kfree(current->pi_state_cache);
852 * Make sure we are holding no locks:
854 debug_check_no_locks_held();
857 exit_io_context(tsk);
859 if (tsk->splice_pipe)
860 free_pipe_info(tsk->splice_pipe);
862 if (tsk->task_frag.page)
863 put_page(tsk->task_frag.page);
865 validate_creds_for_do_exit(tsk);
870 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
872 exit_tasks_rcu_finish();
874 lockdep_free_task(tsk);
877 EXPORT_SYMBOL_GPL(do_exit);
879 void complete_and_exit(struct completion *comp, long code)
886 EXPORT_SYMBOL(complete_and_exit);
888 SYSCALL_DEFINE1(exit, int, error_code)
890 do_exit((error_code&0xff)<<8);
894 * Take down every thread in the group. This is called by fatal signals
895 * as well as by sys_exit_group (below).
898 do_group_exit(int exit_code)
900 struct signal_struct *sig = current->signal;
902 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
904 if (signal_group_exit(sig))
905 exit_code = sig->group_exit_code;
906 else if (!thread_group_empty(current)) {
907 struct sighand_struct *const sighand = current->sighand;
909 spin_lock_irq(&sighand->siglock);
910 if (signal_group_exit(sig))
911 /* Another thread got here before we took the lock. */
912 exit_code = sig->group_exit_code;
914 sig->group_exit_code = exit_code;
915 sig->flags = SIGNAL_GROUP_EXIT;
916 zap_other_threads(current);
918 spin_unlock_irq(&sighand->siglock);
926 * this kills every thread in the thread group. Note that any externally
927 * wait4()-ing process will get the correct exit code - even if this
928 * thread is not the thread group leader.
930 SYSCALL_DEFINE1(exit_group, int, error_code)
932 do_group_exit((error_code & 0xff) << 8);
945 enum pid_type wo_type;
949 struct waitid_info *wo_info;
951 struct rusage *wo_rusage;
953 wait_queue_entry_t child_wait;
957 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
959 return wo->wo_type == PIDTYPE_MAX ||
960 task_pid_type(p, wo->wo_type) == wo->wo_pid;
964 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
966 if (!eligible_pid(wo, p))
970 * Wait for all children (clone and not) if __WALL is set or
971 * if it is traced by us.
973 if (ptrace || (wo->wo_flags & __WALL))
977 * Otherwise, wait for clone children *only* if __WCLONE is set;
978 * otherwise, wait for non-clone children *only*.
980 * Note: a "clone" child here is one that reports to its parent
981 * using a signal other than SIGCHLD, or a non-leader thread which
982 * we can only see if it is traced by us.
984 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
991 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
992 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
993 * the lock and this task is uninteresting. If we return nonzero, we have
994 * released the lock and the system call should return.
996 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
999 pid_t pid = task_pid_vnr(p);
1000 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1001 struct waitid_info *infop;
1003 if (!likely(wo->wo_flags & WEXITED))
1006 if (unlikely(wo->wo_flags & WNOWAIT)) {
1007 status = p->exit_code;
1009 read_unlock(&tasklist_lock);
1010 sched_annotate_sleep();
1012 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1017 * Move the task's state to DEAD/TRACE, only one thread can do this.
1019 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1020 EXIT_TRACE : EXIT_DEAD;
1021 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1024 * We own this thread, nobody else can reap it.
1026 read_unlock(&tasklist_lock);
1027 sched_annotate_sleep();
1030 * Check thread_group_leader() to exclude the traced sub-threads.
1032 if (state == EXIT_DEAD && thread_group_leader(p)) {
1033 struct signal_struct *sig = p->signal;
1034 struct signal_struct *psig = current->signal;
1035 unsigned long maxrss;
1036 u64 tgutime, tgstime;
1039 * The resource counters for the group leader are in its
1040 * own task_struct. Those for dead threads in the group
1041 * are in its signal_struct, as are those for the child
1042 * processes it has previously reaped. All these
1043 * accumulate in the parent's signal_struct c* fields.
1045 * We don't bother to take a lock here to protect these
1046 * p->signal fields because the whole thread group is dead
1047 * and nobody can change them.
1049 * psig->stats_lock also protects us from our sub-theads
1050 * which can reap other children at the same time. Until
1051 * we change k_getrusage()-like users to rely on this lock
1052 * we have to take ->siglock as well.
1054 * We use thread_group_cputime_adjusted() to get times for
1055 * the thread group, which consolidates times for all threads
1056 * in the group including the group leader.
1058 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1059 spin_lock_irq(¤t->sighand->siglock);
1060 write_seqlock(&psig->stats_lock);
1061 psig->cutime += tgutime + sig->cutime;
1062 psig->cstime += tgstime + sig->cstime;
1063 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1065 p->min_flt + sig->min_flt + sig->cmin_flt;
1067 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1069 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1071 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1073 task_io_get_inblock(p) +
1074 sig->inblock + sig->cinblock;
1076 task_io_get_oublock(p) +
1077 sig->oublock + sig->coublock;
1078 maxrss = max(sig->maxrss, sig->cmaxrss);
1079 if (psig->cmaxrss < maxrss)
1080 psig->cmaxrss = maxrss;
1081 task_io_accounting_add(&psig->ioac, &p->ioac);
1082 task_io_accounting_add(&psig->ioac, &sig->ioac);
1083 write_sequnlock(&psig->stats_lock);
1084 spin_unlock_irq(¤t->sighand->siglock);
1088 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1089 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1090 ? p->signal->group_exit_code : p->exit_code;
1091 wo->wo_stat = status;
1093 if (state == EXIT_TRACE) {
1094 write_lock_irq(&tasklist_lock);
1095 /* We dropped tasklist, ptracer could die and untrace */
1098 /* If parent wants a zombie, don't release it now */
1099 state = EXIT_ZOMBIE;
1100 if (do_notify_parent(p, p->exit_signal))
1102 p->exit_state = state;
1103 write_unlock_irq(&tasklist_lock);
1105 if (state == EXIT_DEAD)
1109 infop = wo->wo_info;
1111 if ((status & 0x7f) == 0) {
1112 infop->cause = CLD_EXITED;
1113 infop->status = status >> 8;
1115 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1116 infop->status = status & 0x7f;
1125 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1128 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1129 return &p->exit_code;
1131 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1132 return &p->signal->group_exit_code;
1138 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1140 * @ptrace: is the wait for ptrace
1141 * @p: task to wait for
1143 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1146 * read_lock(&tasklist_lock), which is released if return value is
1147 * non-zero. Also, grabs and releases @p->sighand->siglock.
1150 * 0 if wait condition didn't exist and search for other wait conditions
1151 * should continue. Non-zero return, -errno on failure and @p's pid on
1152 * success, implies that tasklist_lock is released and wait condition
1153 * search should terminate.
1155 static int wait_task_stopped(struct wait_opts *wo,
1156 int ptrace, struct task_struct *p)
1158 struct waitid_info *infop;
1159 int exit_code, *p_code, why;
1160 uid_t uid = 0; /* unneeded, required by compiler */
1164 * Traditionally we see ptrace'd stopped tasks regardless of options.
1166 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1169 if (!task_stopped_code(p, ptrace))
1173 spin_lock_irq(&p->sighand->siglock);
1175 p_code = task_stopped_code(p, ptrace);
1176 if (unlikely(!p_code))
1179 exit_code = *p_code;
1183 if (!unlikely(wo->wo_flags & WNOWAIT))
1186 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1188 spin_unlock_irq(&p->sighand->siglock);
1193 * Now we are pretty sure this task is interesting.
1194 * Make sure it doesn't get reaped out from under us while we
1195 * give up the lock and then examine it below. We don't want to
1196 * keep holding onto the tasklist_lock while we call getrusage and
1197 * possibly take page faults for user memory.
1200 pid = task_pid_vnr(p);
1201 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1202 read_unlock(&tasklist_lock);
1203 sched_annotate_sleep();
1205 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1208 if (likely(!(wo->wo_flags & WNOWAIT)))
1209 wo->wo_stat = (exit_code << 8) | 0x7f;
1211 infop = wo->wo_info;
1214 infop->status = exit_code;
1222 * Handle do_wait work for one task in a live, non-stopped state.
1223 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1224 * the lock and this task is uninteresting. If we return nonzero, we have
1225 * released the lock and the system call should return.
1227 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1229 struct waitid_info *infop;
1233 if (!unlikely(wo->wo_flags & WCONTINUED))
1236 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1239 spin_lock_irq(&p->sighand->siglock);
1240 /* Re-check with the lock held. */
1241 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1242 spin_unlock_irq(&p->sighand->siglock);
1245 if (!unlikely(wo->wo_flags & WNOWAIT))
1246 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1247 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1248 spin_unlock_irq(&p->sighand->siglock);
1250 pid = task_pid_vnr(p);
1252 read_unlock(&tasklist_lock);
1253 sched_annotate_sleep();
1255 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1258 infop = wo->wo_info;
1260 wo->wo_stat = 0xffff;
1262 infop->cause = CLD_CONTINUED;
1265 infop->status = SIGCONT;
1271 * Consider @p for a wait by @parent.
1273 * -ECHILD should be in ->notask_error before the first call.
1274 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1275 * Returns zero if the search for a child should continue;
1276 * then ->notask_error is 0 if @p is an eligible child,
1279 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1280 struct task_struct *p)
1283 * We can race with wait_task_zombie() from another thread.
1284 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1285 * can't confuse the checks below.
1287 int exit_state = READ_ONCE(p->exit_state);
1290 if (unlikely(exit_state == EXIT_DEAD))
1293 ret = eligible_child(wo, ptrace, p);
1297 if (unlikely(exit_state == EXIT_TRACE)) {
1299 * ptrace == 0 means we are the natural parent. In this case
1300 * we should clear notask_error, debugger will notify us.
1302 if (likely(!ptrace))
1303 wo->notask_error = 0;
1307 if (likely(!ptrace) && unlikely(p->ptrace)) {
1309 * If it is traced by its real parent's group, just pretend
1310 * the caller is ptrace_do_wait() and reap this child if it
1313 * This also hides group stop state from real parent; otherwise
1314 * a single stop can be reported twice as group and ptrace stop.
1315 * If a ptracer wants to distinguish these two events for its
1316 * own children it should create a separate process which takes
1317 * the role of real parent.
1319 if (!ptrace_reparented(p))
1324 if (exit_state == EXIT_ZOMBIE) {
1325 /* we don't reap group leaders with subthreads */
1326 if (!delay_group_leader(p)) {
1328 * A zombie ptracee is only visible to its ptracer.
1329 * Notification and reaping will be cascaded to the
1330 * real parent when the ptracer detaches.
1332 if (unlikely(ptrace) || likely(!p->ptrace))
1333 return wait_task_zombie(wo, p);
1337 * Allow access to stopped/continued state via zombie by
1338 * falling through. Clearing of notask_error is complex.
1342 * If WEXITED is set, notask_error should naturally be
1343 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1344 * so, if there are live subthreads, there are events to
1345 * wait for. If all subthreads are dead, it's still safe
1346 * to clear - this function will be called again in finite
1347 * amount time once all the subthreads are released and
1348 * will then return without clearing.
1352 * Stopped state is per-task and thus can't change once the
1353 * target task dies. Only continued and exited can happen.
1354 * Clear notask_error if WCONTINUED | WEXITED.
1356 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1357 wo->notask_error = 0;
1360 * @p is alive and it's gonna stop, continue or exit, so
1361 * there always is something to wait for.
1363 wo->notask_error = 0;
1367 * Wait for stopped. Depending on @ptrace, different stopped state
1368 * is used and the two don't interact with each other.
1370 ret = wait_task_stopped(wo, ptrace, p);
1375 * Wait for continued. There's only one continued state and the
1376 * ptracer can consume it which can confuse the real parent. Don't
1377 * use WCONTINUED from ptracer. You don't need or want it.
1379 return wait_task_continued(wo, p);
1383 * Do the work of do_wait() for one thread in the group, @tsk.
1385 * -ECHILD should be in ->notask_error before the first call.
1386 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1387 * Returns zero if the search for a child should continue; then
1388 * ->notask_error is 0 if there were any eligible children,
1391 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1393 struct task_struct *p;
1395 list_for_each_entry(p, &tsk->children, sibling) {
1396 int ret = wait_consider_task(wo, 0, p);
1405 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1407 struct task_struct *p;
1409 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1410 int ret = wait_consider_task(wo, 1, p);
1419 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1420 int sync, void *key)
1422 struct wait_opts *wo = container_of(wait, struct wait_opts,
1424 struct task_struct *p = key;
1426 if (!eligible_pid(wo, p))
1429 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1432 return default_wake_function(wait, mode, sync, key);
1435 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1437 __wake_up_sync_key(&parent->signal->wait_chldexit,
1438 TASK_INTERRUPTIBLE, p);
1441 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1442 struct task_struct *target)
1444 struct task_struct *parent =
1445 !ptrace ? target->real_parent : target->parent;
1447 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1448 same_thread_group(current, parent));
1452 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1453 * and tracee lists to find the target task.
1455 static int do_wait_pid(struct wait_opts *wo)
1458 struct task_struct *target;
1462 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1463 if (target && is_effectively_child(wo, ptrace, target)) {
1464 retval = wait_consider_task(wo, ptrace, target);
1470 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1471 if (target && target->ptrace &&
1472 is_effectively_child(wo, ptrace, target)) {
1473 retval = wait_consider_task(wo, ptrace, target);
1481 static long do_wait(struct wait_opts *wo)
1485 trace_sched_process_wait(wo->wo_pid);
1487 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1488 wo->child_wait.private = current;
1489 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1492 * If there is nothing that can match our criteria, just get out.
1493 * We will clear ->notask_error to zero if we see any child that
1494 * might later match our criteria, even if we are not able to reap
1497 wo->notask_error = -ECHILD;
1498 if ((wo->wo_type < PIDTYPE_MAX) &&
1499 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1502 set_current_state(TASK_INTERRUPTIBLE);
1503 read_lock(&tasklist_lock);
1505 if (wo->wo_type == PIDTYPE_PID) {
1506 retval = do_wait_pid(wo);
1510 struct task_struct *tsk = current;
1513 retval = do_wait_thread(wo, tsk);
1517 retval = ptrace_do_wait(wo, tsk);
1521 if (wo->wo_flags & __WNOTHREAD)
1523 } while_each_thread(current, tsk);
1525 read_unlock(&tasklist_lock);
1528 retval = wo->notask_error;
1529 if (!retval && !(wo->wo_flags & WNOHANG)) {
1530 retval = -ERESTARTSYS;
1531 if (!signal_pending(current)) {
1537 __set_current_state(TASK_RUNNING);
1538 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1542 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1543 int options, struct rusage *ru)
1545 struct wait_opts wo;
1546 struct pid *pid = NULL;
1549 unsigned int f_flags = 0;
1551 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1552 __WNOTHREAD|__WCLONE|__WALL))
1554 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1566 pid = find_get_pid(upid);
1569 type = PIDTYPE_PGID;
1574 pid = find_get_pid(upid);
1576 pid = get_task_pid(current, PIDTYPE_PGID);
1583 pid = pidfd_get_pid(upid, &f_flags);
1585 return PTR_ERR(pid);
1594 wo.wo_flags = options;
1597 if (f_flags & O_NONBLOCK)
1598 wo.wo_flags |= WNOHANG;
1601 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1608 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1609 infop, int, options, struct rusage __user *, ru)
1612 struct waitid_info info = {.status = 0};
1613 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1619 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1625 if (!user_write_access_begin(infop, sizeof(*infop)))
1628 unsafe_put_user(signo, &infop->si_signo, Efault);
1629 unsafe_put_user(0, &infop->si_errno, Efault);
1630 unsafe_put_user(info.cause, &infop->si_code, Efault);
1631 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1632 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1633 unsafe_put_user(info.status, &infop->si_status, Efault);
1634 user_write_access_end();
1637 user_write_access_end();
1641 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1644 struct wait_opts wo;
1645 struct pid *pid = NULL;
1649 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1650 __WNOTHREAD|__WCLONE|__WALL))
1653 /* -INT_MIN is not defined */
1654 if (upid == INT_MIN)
1659 else if (upid < 0) {
1660 type = PIDTYPE_PGID;
1661 pid = find_get_pid(-upid);
1662 } else if (upid == 0) {
1663 type = PIDTYPE_PGID;
1664 pid = get_task_pid(current, PIDTYPE_PGID);
1665 } else /* upid > 0 */ {
1667 pid = find_get_pid(upid);
1672 wo.wo_flags = options | WEXITED;
1678 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1684 int kernel_wait(pid_t pid, int *stat)
1686 struct wait_opts wo = {
1687 .wo_type = PIDTYPE_PID,
1688 .wo_pid = find_get_pid(pid),
1689 .wo_flags = WEXITED,
1694 if (ret > 0 && wo.wo_stat)
1700 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1701 int, options, struct rusage __user *, ru)
1704 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1707 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1713 #ifdef __ARCH_WANT_SYS_WAITPID
1716 * sys_waitpid() remains for compatibility. waitpid() should be
1717 * implemented by calling sys_wait4() from libc.a.
1719 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1721 return kernel_wait4(pid, stat_addr, options, NULL);
1726 #ifdef CONFIG_COMPAT
1727 COMPAT_SYSCALL_DEFINE4(wait4,
1729 compat_uint_t __user *, stat_addr,
1731 struct compat_rusage __user *, ru)
1734 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1736 if (ru && put_compat_rusage(&r, ru))
1742 COMPAT_SYSCALL_DEFINE5(waitid,
1743 int, which, compat_pid_t, pid,
1744 struct compat_siginfo __user *, infop, int, options,
1745 struct compat_rusage __user *, uru)
1748 struct waitid_info info = {.status = 0};
1749 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1755 /* kernel_waitid() overwrites everything in ru */
1756 if (COMPAT_USE_64BIT_TIME)
1757 err = copy_to_user(uru, &ru, sizeof(ru));
1759 err = put_compat_rusage(&ru, uru);
1768 if (!user_write_access_begin(infop, sizeof(*infop)))
1771 unsafe_put_user(signo, &infop->si_signo, Efault);
1772 unsafe_put_user(0, &infop->si_errno, Efault);
1773 unsafe_put_user(info.cause, &infop->si_code, Efault);
1774 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1775 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1776 unsafe_put_user(info.status, &infop->si_status, Efault);
1777 user_write_access_end();
1780 user_write_access_end();
1786 * thread_group_exited - check that a thread group has exited
1787 * @pid: tgid of thread group to be checked.
1789 * Test if the thread group represented by tgid has exited (all
1790 * threads are zombies, dead or completely gone).
1792 * Return: true if the thread group has exited. false otherwise.
1794 bool thread_group_exited(struct pid *pid)
1796 struct task_struct *task;
1800 task = pid_task(pid, PIDTYPE_PID);
1802 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1807 EXPORT_SYMBOL(thread_group_exited);
1809 __weak void abort(void)
1813 /* if that doesn't kill us, halt */
1814 panic("Oops failed to kill thread");
1816 EXPORT_SYMBOL(abort);