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>
66 #include <linux/kprobes.h>
67 #include <linux/rethook.h>
69 #include <linux/uaccess.h>
70 #include <asm/unistd.h>
71 #include <asm/mmu_context.h>
73 static void __unhash_process(struct task_struct *p, bool group_dead)
76 detach_pid(p, PIDTYPE_PID);
78 detach_pid(p, PIDTYPE_TGID);
79 detach_pid(p, PIDTYPE_PGID);
80 detach_pid(p, PIDTYPE_SID);
82 list_del_rcu(&p->tasks);
83 list_del_init(&p->sibling);
84 __this_cpu_dec(process_counts);
86 list_del_rcu(&p->thread_group);
87 list_del_rcu(&p->thread_node);
91 * This function expects the tasklist_lock write-locked.
93 static void __exit_signal(struct task_struct *tsk)
95 struct signal_struct *sig = tsk->signal;
96 bool group_dead = thread_group_leader(tsk);
97 struct sighand_struct *sighand;
98 struct tty_struct *tty;
101 sighand = rcu_dereference_check(tsk->sighand,
102 lockdep_tasklist_lock_is_held());
103 spin_lock(&sighand->siglock);
105 #ifdef CONFIG_POSIX_TIMERS
106 posix_cpu_timers_exit(tsk);
108 posix_cpu_timers_exit_group(tsk);
116 * If there is any task waiting for the group exit
119 if (sig->notify_count > 0 && !--sig->notify_count)
120 wake_up_process(sig->group_exec_task);
122 if (tsk == sig->curr_target)
123 sig->curr_target = next_thread(tsk);
126 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
127 sizeof(unsigned long long));
130 * Accumulate here the counters for all threads as they die. We could
131 * skip the group leader because it is the last user of signal_struct,
132 * but we want to avoid the race with thread_group_cputime() which can
133 * see the empty ->thread_head list.
135 task_cputime(tsk, &utime, &stime);
136 write_seqlock(&sig->stats_lock);
139 sig->gtime += task_gtime(tsk);
140 sig->min_flt += tsk->min_flt;
141 sig->maj_flt += tsk->maj_flt;
142 sig->nvcsw += tsk->nvcsw;
143 sig->nivcsw += tsk->nivcsw;
144 sig->inblock += task_io_get_inblock(tsk);
145 sig->oublock += task_io_get_oublock(tsk);
146 task_io_accounting_add(&sig->ioac, &tsk->ioac);
147 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
149 __unhash_process(tsk, group_dead);
150 write_sequnlock(&sig->stats_lock);
153 * Do this under ->siglock, we can race with another thread
154 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
156 flush_sigqueue(&tsk->pending);
158 spin_unlock(&sighand->siglock);
160 __cleanup_sighand(sighand);
161 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
163 flush_sigqueue(&sig->shared_pending);
168 static void delayed_put_task_struct(struct rcu_head *rhp)
170 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
172 kprobe_flush_task(tsk);
173 rethook_flush_task(tsk);
174 perf_event_delayed_put(tsk);
175 trace_sched_process_free(tsk);
176 put_task_struct(tsk);
179 void put_task_struct_rcu_user(struct task_struct *task)
181 if (refcount_dec_and_test(&task->rcu_users))
182 call_rcu(&task->rcu, delayed_put_task_struct);
185 void release_task(struct task_struct *p)
187 struct task_struct *leader;
188 struct pid *thread_pid;
191 /* don't need to get the RCU readlock here - the process is dead and
192 * can't be modifying its own credentials. But shut RCU-lockdep up */
194 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
199 write_lock_irq(&tasklist_lock);
200 ptrace_release_task(p);
201 thread_pid = get_pid(p->thread_pid);
205 * If we are the last non-leader member of the thread
206 * group, and the leader is zombie, then notify the
207 * group leader's parent process. (if it wants notification.)
210 leader = p->group_leader;
211 if (leader != p && thread_group_empty(leader)
212 && leader->exit_state == EXIT_ZOMBIE) {
214 * If we were the last child thread and the leader has
215 * exited already, and the leader's parent ignores SIGCHLD,
216 * then we are the one who should release the leader.
218 zap_leader = do_notify_parent(leader, leader->exit_signal);
220 leader->exit_state = EXIT_DEAD;
223 write_unlock_irq(&tasklist_lock);
224 seccomp_filter_release(p);
225 proc_flush_pid(thread_pid);
228 put_task_struct_rcu_user(p);
231 if (unlikely(zap_leader))
235 int rcuwait_wake_up(struct rcuwait *w)
238 struct task_struct *task;
243 * Order condition vs @task, such that everything prior to the load
244 * of @task is visible. This is the condition as to why the user called
245 * rcuwait_wake() in the first place. Pairs with set_current_state()
246 * barrier (A) in rcuwait_wait_event().
249 * [S] tsk = current [S] cond = true
255 task = rcu_dereference(w->task);
257 ret = wake_up_process(task);
262 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
265 * Determine if a process group is "orphaned", according to the POSIX
266 * definition in 2.2.2.52. Orphaned process groups are not to be affected
267 * by terminal-generated stop signals. Newly orphaned process groups are
268 * to receive a SIGHUP and a SIGCONT.
270 * "I ask you, have you ever known what it is to be an orphan?"
272 static int will_become_orphaned_pgrp(struct pid *pgrp,
273 struct task_struct *ignored_task)
275 struct task_struct *p;
277 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
278 if ((p == ignored_task) ||
279 (p->exit_state && thread_group_empty(p)) ||
280 is_global_init(p->real_parent))
283 if (task_pgrp(p->real_parent) != pgrp &&
284 task_session(p->real_parent) == task_session(p))
286 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
291 int is_current_pgrp_orphaned(void)
295 read_lock(&tasklist_lock);
296 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
297 read_unlock(&tasklist_lock);
302 static bool has_stopped_jobs(struct pid *pgrp)
304 struct task_struct *p;
306 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
307 if (p->signal->flags & SIGNAL_STOP_STOPPED)
309 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
315 * Check to see if any process groups have become orphaned as
316 * a result of our exiting, and if they have any stopped jobs,
317 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
320 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
322 struct pid *pgrp = task_pgrp(tsk);
323 struct task_struct *ignored_task = tsk;
326 /* exit: our father is in a different pgrp than
327 * we are and we were the only connection outside.
329 parent = tsk->real_parent;
331 /* reparent: our child is in a different pgrp than
332 * we are, and it was the only connection outside.
336 if (task_pgrp(parent) != pgrp &&
337 task_session(parent) == task_session(tsk) &&
338 will_become_orphaned_pgrp(pgrp, ignored_task) &&
339 has_stopped_jobs(pgrp)) {
340 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
341 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
345 static void coredump_task_exit(struct task_struct *tsk)
347 struct core_state *core_state;
350 * Serialize with any possible pending coredump.
351 * We must hold siglock around checking core_state
352 * and setting PF_POSTCOREDUMP. The core-inducing thread
353 * will increment ->nr_threads for each thread in the
354 * group without PF_POSTCOREDUMP set.
356 spin_lock_irq(&tsk->sighand->siglock);
357 tsk->flags |= PF_POSTCOREDUMP;
358 core_state = tsk->signal->core_state;
359 spin_unlock_irq(&tsk->sighand->siglock);
361 struct core_thread self;
364 if (self.task->flags & PF_SIGNALED)
365 self.next = xchg(&core_state->dumper.next, &self);
369 * Implies mb(), the result of xchg() must be visible
370 * to core_state->dumper.
372 if (atomic_dec_and_test(&core_state->nr_threads))
373 complete(&core_state->startup);
376 set_current_state(TASK_UNINTERRUPTIBLE);
377 if (!self.task) /* see coredump_finish() */
379 freezable_schedule();
381 __set_current_state(TASK_RUNNING);
387 * A task is exiting. If it owned this mm, find a new owner for the mm.
389 void mm_update_next_owner(struct mm_struct *mm)
391 struct task_struct *c, *g, *p = current;
395 * If the exiting or execing task is not the owner, it's
396 * someone else's problem.
401 * The current owner is exiting/execing and there are no other
402 * candidates. Do not leave the mm pointing to a possibly
403 * freed task structure.
405 if (atomic_read(&mm->mm_users) <= 1) {
406 WRITE_ONCE(mm->owner, NULL);
410 read_lock(&tasklist_lock);
412 * Search in the children
414 list_for_each_entry(c, &p->children, sibling) {
416 goto assign_new_owner;
420 * Search in the siblings
422 list_for_each_entry(c, &p->real_parent->children, sibling) {
424 goto assign_new_owner;
428 * Search through everything else, we should not get here often.
430 for_each_process(g) {
431 if (g->flags & PF_KTHREAD)
433 for_each_thread(g, c) {
435 goto assign_new_owner;
440 read_unlock(&tasklist_lock);
442 * We found no owner yet mm_users > 1: this implies that we are
443 * most likely racing with swapoff (try_to_unuse()) or /proc or
444 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
446 WRITE_ONCE(mm->owner, NULL);
453 * The task_lock protects c->mm from changing.
454 * We always want mm->owner->mm == mm
458 * Delay read_unlock() till we have the task_lock()
459 * to ensure that c does not slip away underneath us
461 read_unlock(&tasklist_lock);
467 WRITE_ONCE(mm->owner, c);
471 #endif /* CONFIG_MEMCG */
474 * Turn us into a lazy TLB process if we
477 static void exit_mm(void)
479 struct mm_struct *mm = current->mm;
481 exit_mm_release(current, mm);
487 BUG_ON(mm != current->active_mm);
488 /* more a memory barrier than a real lock */
491 * When a thread stops operating on an address space, the loop
492 * in membarrier_private_expedited() may not observe that
493 * tsk->mm, and the loop in membarrier_global_expedited() may
494 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
495 * rq->membarrier_state, so those would not issue an IPI.
496 * Membarrier requires a memory barrier after accessing
497 * user-space memory, before clearing tsk->mm or the
498 * rq->membarrier_state.
500 smp_mb__after_spinlock();
503 membarrier_update_current_mm(NULL);
504 enter_lazy_tlb(mm, current);
506 task_unlock(current);
507 mmap_read_unlock(mm);
508 mm_update_next_owner(mm);
510 if (test_thread_flag(TIF_MEMDIE))
514 static struct task_struct *find_alive_thread(struct task_struct *p)
516 struct task_struct *t;
518 for_each_thread(p, t) {
519 if (!(t->flags & PF_EXITING))
525 static struct task_struct *find_child_reaper(struct task_struct *father,
526 struct list_head *dead)
527 __releases(&tasklist_lock)
528 __acquires(&tasklist_lock)
530 struct pid_namespace *pid_ns = task_active_pid_ns(father);
531 struct task_struct *reaper = pid_ns->child_reaper;
532 struct task_struct *p, *n;
534 if (likely(reaper != father))
537 reaper = find_alive_thread(father);
539 pid_ns->child_reaper = reaper;
543 write_unlock_irq(&tasklist_lock);
545 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
546 list_del_init(&p->ptrace_entry);
550 zap_pid_ns_processes(pid_ns);
551 write_lock_irq(&tasklist_lock);
557 * When we die, we re-parent all our children, and try to:
558 * 1. give them to another thread in our thread group, if such a member exists
559 * 2. give it to the first ancestor process which prctl'd itself as a
560 * child_subreaper for its children (like a service manager)
561 * 3. give it to the init process (PID 1) in our pid namespace
563 static struct task_struct *find_new_reaper(struct task_struct *father,
564 struct task_struct *child_reaper)
566 struct task_struct *thread, *reaper;
568 thread = find_alive_thread(father);
572 if (father->signal->has_child_subreaper) {
573 unsigned int ns_level = task_pid(father)->level;
575 * Find the first ->is_child_subreaper ancestor in our pid_ns.
576 * We can't check reaper != child_reaper to ensure we do not
577 * cross the namespaces, the exiting parent could be injected
578 * by setns() + fork().
579 * We check pid->level, this is slightly more efficient than
580 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
582 for (reaper = father->real_parent;
583 task_pid(reaper)->level == ns_level;
584 reaper = reaper->real_parent) {
585 if (reaper == &init_task)
587 if (!reaper->signal->is_child_subreaper)
589 thread = find_alive_thread(reaper);
599 * Any that need to be release_task'd are put on the @dead list.
601 static void reparent_leader(struct task_struct *father, struct task_struct *p,
602 struct list_head *dead)
604 if (unlikely(p->exit_state == EXIT_DEAD))
607 /* We don't want people slaying init. */
608 p->exit_signal = SIGCHLD;
610 /* If it has exited notify the new parent about this child's death. */
612 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
613 if (do_notify_parent(p, p->exit_signal)) {
614 p->exit_state = EXIT_DEAD;
615 list_add(&p->ptrace_entry, dead);
619 kill_orphaned_pgrp(p, father);
623 * This does two things:
625 * A. Make init inherit all the child processes
626 * B. Check to see if any process groups have become orphaned
627 * as a result of our exiting, and if they have any stopped
628 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
630 static void forget_original_parent(struct task_struct *father,
631 struct list_head *dead)
633 struct task_struct *p, *t, *reaper;
635 if (unlikely(!list_empty(&father->ptraced)))
636 exit_ptrace(father, dead);
638 /* Can drop and reacquire tasklist_lock */
639 reaper = find_child_reaper(father, dead);
640 if (list_empty(&father->children))
643 reaper = find_new_reaper(father, reaper);
644 list_for_each_entry(p, &father->children, sibling) {
645 for_each_thread(p, t) {
646 RCU_INIT_POINTER(t->real_parent, reaper);
647 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
648 if (likely(!t->ptrace))
649 t->parent = t->real_parent;
650 if (t->pdeath_signal)
651 group_send_sig_info(t->pdeath_signal,
656 * If this is a threaded reparent there is no need to
657 * notify anyone anything has happened.
659 if (!same_thread_group(reaper, father))
660 reparent_leader(father, p, dead);
662 list_splice_tail_init(&father->children, &reaper->children);
666 * Send signals to all our closest relatives so that they know
667 * to properly mourn us..
669 static void exit_notify(struct task_struct *tsk, int group_dead)
672 struct task_struct *p, *n;
675 write_lock_irq(&tasklist_lock);
676 forget_original_parent(tsk, &dead);
679 kill_orphaned_pgrp(tsk->group_leader, NULL);
681 tsk->exit_state = EXIT_ZOMBIE;
682 if (unlikely(tsk->ptrace)) {
683 int sig = thread_group_leader(tsk) &&
684 thread_group_empty(tsk) &&
685 !ptrace_reparented(tsk) ?
686 tsk->exit_signal : SIGCHLD;
687 autoreap = do_notify_parent(tsk, sig);
688 } else if (thread_group_leader(tsk)) {
689 autoreap = thread_group_empty(tsk) &&
690 do_notify_parent(tsk, tsk->exit_signal);
696 tsk->exit_state = EXIT_DEAD;
697 list_add(&tsk->ptrace_entry, &dead);
700 /* mt-exec, de_thread() is waiting for group leader */
701 if (unlikely(tsk->signal->notify_count < 0))
702 wake_up_process(tsk->signal->group_exec_task);
703 write_unlock_irq(&tasklist_lock);
705 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
706 list_del_init(&p->ptrace_entry);
711 #ifdef CONFIG_DEBUG_STACK_USAGE
712 static void check_stack_usage(void)
714 static DEFINE_SPINLOCK(low_water_lock);
715 static int lowest_to_date = THREAD_SIZE;
718 free = stack_not_used(current);
720 if (free >= lowest_to_date)
723 spin_lock(&low_water_lock);
724 if (free < lowest_to_date) {
725 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
726 current->comm, task_pid_nr(current), free);
727 lowest_to_date = free;
729 spin_unlock(&low_water_lock);
732 static inline void check_stack_usage(void) {}
735 void __noreturn do_exit(long code)
737 struct task_struct *tsk = current;
744 coredump_task_exit(tsk);
745 ptrace_event(PTRACE_EVENT_EXIT, code);
747 validate_creds_for_do_exit(tsk);
749 io_uring_files_cancel();
750 exit_signals(tsk); /* sets PF_EXITING */
752 /* sync mm's RSS info before statistics gathering */
754 sync_mm_rss(tsk->mm);
755 acct_update_integrals(tsk);
756 group_dead = atomic_dec_and_test(&tsk->signal->live);
759 * If the last thread of global init has exited, panic
760 * immediately to get a useable coredump.
762 if (unlikely(is_global_init(tsk)))
763 panic("Attempted to kill init! exitcode=0x%08x\n",
764 tsk->signal->group_exit_code ?: (int)code);
766 #ifdef CONFIG_POSIX_TIMERS
767 hrtimer_cancel(&tsk->signal->real_timer);
768 exit_itimers(tsk->signal);
771 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
773 acct_collect(code, group_dead);
778 tsk->exit_code = code;
779 taskstats_exit(tsk, group_dead);
785 trace_sched_process_exit(tsk);
792 disassociate_ctty(1);
793 exit_task_namespaces(tsk);
798 * Flush inherited counters to the parent - before the parent
799 * gets woken up by child-exit notifications.
801 * because of cgroup mode, must be called before cgroup_exit()
803 perf_event_exit_task(tsk);
805 sched_autogroup_exit_task(tsk);
809 * FIXME: do that only when needed, using sched_exit tracepoint
811 flush_ptrace_hw_breakpoint(tsk);
813 exit_tasks_rcu_start();
814 exit_notify(tsk, group_dead);
815 proc_exit_connector(tsk);
816 mpol_put_task_policy(tsk);
818 if (unlikely(current->pi_state_cache))
819 kfree(current->pi_state_cache);
822 * Make sure we are holding no locks:
824 debug_check_no_locks_held();
827 exit_io_context(tsk);
829 if (tsk->splice_pipe)
830 free_pipe_info(tsk->splice_pipe);
832 if (tsk->task_frag.page)
833 put_page(tsk->task_frag.page);
835 validate_creds_for_do_exit(tsk);
836 exit_task_stack_account(tsk);
841 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
843 exit_tasks_rcu_finish();
845 lockdep_free_task(tsk);
849 void __noreturn make_task_dead(int signr)
852 * Take the task off the cpu after something catastrophic has
855 * We can get here from a kernel oops, sometimes with preemption off.
856 * Start by checking for critical errors.
857 * Then fix up important state like USER_DS and preemption.
858 * Then do everything else.
860 struct task_struct *tsk = current;
862 if (unlikely(in_interrupt()))
863 panic("Aiee, killing interrupt handler!");
864 if (unlikely(!tsk->pid))
865 panic("Attempted to kill the idle task!");
867 if (unlikely(in_atomic())) {
868 pr_info("note: %s[%d] exited with preempt_count %d\n",
869 current->comm, task_pid_nr(current),
871 preempt_count_set(PREEMPT_ENABLED);
875 * We're taking recursive faults here in make_task_dead. Safest is to just
876 * leave this task alone and wait for reboot.
878 if (unlikely(tsk->flags & PF_EXITING)) {
879 pr_alert("Fixing recursive fault but reboot is needed!\n");
880 futex_exit_recursive(tsk);
881 tsk->exit_state = EXIT_DEAD;
882 refcount_inc(&tsk->rcu_users);
889 SYSCALL_DEFINE1(exit, int, error_code)
891 do_exit((error_code&0xff)<<8);
895 * Take down every thread in the group. This is called by fatal signals
896 * as well as by sys_exit_group (below).
899 do_group_exit(int exit_code)
901 struct signal_struct *sig = current->signal;
903 if (sig->flags & SIGNAL_GROUP_EXIT)
904 exit_code = sig->group_exit_code;
905 else if (sig->group_exec_task)
907 else if (!thread_group_empty(current)) {
908 struct sighand_struct *const sighand = current->sighand;
910 spin_lock_irq(&sighand->siglock);
911 if (sig->flags & SIGNAL_GROUP_EXIT)
912 /* Another thread got here before we took the lock. */
913 exit_code = sig->group_exit_code;
914 else if (sig->group_exec_task)
917 sig->group_exit_code = exit_code;
918 sig->flags = SIGNAL_GROUP_EXIT;
919 zap_other_threads(current);
921 spin_unlock_irq(&sighand->siglock);
929 * this kills every thread in the thread group. Note that any externally
930 * wait4()-ing process will get the correct exit code - even if this
931 * thread is not the thread group leader.
933 SYSCALL_DEFINE1(exit_group, int, error_code)
935 do_group_exit((error_code & 0xff) << 8);
948 enum pid_type wo_type;
952 struct waitid_info *wo_info;
954 struct rusage *wo_rusage;
956 wait_queue_entry_t child_wait;
960 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
962 return wo->wo_type == PIDTYPE_MAX ||
963 task_pid_type(p, wo->wo_type) == wo->wo_pid;
967 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
969 if (!eligible_pid(wo, p))
973 * Wait for all children (clone and not) if __WALL is set or
974 * if it is traced by us.
976 if (ptrace || (wo->wo_flags & __WALL))
980 * Otherwise, wait for clone children *only* if __WCLONE is set;
981 * otherwise, wait for non-clone children *only*.
983 * Note: a "clone" child here is one that reports to its parent
984 * using a signal other than SIGCHLD, or a non-leader thread which
985 * we can only see if it is traced by us.
987 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
994 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
995 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
996 * the lock and this task is uninteresting. If we return nonzero, we have
997 * released the lock and the system call should return.
999 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1002 pid_t pid = task_pid_vnr(p);
1003 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1004 struct waitid_info *infop;
1006 if (!likely(wo->wo_flags & WEXITED))
1009 if (unlikely(wo->wo_flags & WNOWAIT)) {
1010 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1011 ? p->signal->group_exit_code : p->exit_code;
1013 read_unlock(&tasklist_lock);
1014 sched_annotate_sleep();
1016 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1021 * Move the task's state to DEAD/TRACE, only one thread can do this.
1023 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1024 EXIT_TRACE : EXIT_DEAD;
1025 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1028 * We own this thread, nobody else can reap it.
1030 read_unlock(&tasklist_lock);
1031 sched_annotate_sleep();
1034 * Check thread_group_leader() to exclude the traced sub-threads.
1036 if (state == EXIT_DEAD && thread_group_leader(p)) {
1037 struct signal_struct *sig = p->signal;
1038 struct signal_struct *psig = current->signal;
1039 unsigned long maxrss;
1040 u64 tgutime, tgstime;
1043 * The resource counters for the group leader are in its
1044 * own task_struct. Those for dead threads in the group
1045 * are in its signal_struct, as are those for the child
1046 * processes it has previously reaped. All these
1047 * accumulate in the parent's signal_struct c* fields.
1049 * We don't bother to take a lock here to protect these
1050 * p->signal fields because the whole thread group is dead
1051 * and nobody can change them.
1053 * psig->stats_lock also protects us from our sub-theads
1054 * which can reap other children at the same time. Until
1055 * we change k_getrusage()-like users to rely on this lock
1056 * we have to take ->siglock as well.
1058 * We use thread_group_cputime_adjusted() to get times for
1059 * the thread group, which consolidates times for all threads
1060 * in the group including the group leader.
1062 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1063 spin_lock_irq(¤t->sighand->siglock);
1064 write_seqlock(&psig->stats_lock);
1065 psig->cutime += tgutime + sig->cutime;
1066 psig->cstime += tgstime + sig->cstime;
1067 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1069 p->min_flt + sig->min_flt + sig->cmin_flt;
1071 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1073 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1075 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1077 task_io_get_inblock(p) +
1078 sig->inblock + sig->cinblock;
1080 task_io_get_oublock(p) +
1081 sig->oublock + sig->coublock;
1082 maxrss = max(sig->maxrss, sig->cmaxrss);
1083 if (psig->cmaxrss < maxrss)
1084 psig->cmaxrss = maxrss;
1085 task_io_accounting_add(&psig->ioac, &p->ioac);
1086 task_io_accounting_add(&psig->ioac, &sig->ioac);
1087 write_sequnlock(&psig->stats_lock);
1088 spin_unlock_irq(¤t->sighand->siglock);
1092 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1093 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1094 ? p->signal->group_exit_code : p->exit_code;
1095 wo->wo_stat = status;
1097 if (state == EXIT_TRACE) {
1098 write_lock_irq(&tasklist_lock);
1099 /* We dropped tasklist, ptracer could die and untrace */
1102 /* If parent wants a zombie, don't release it now */
1103 state = EXIT_ZOMBIE;
1104 if (do_notify_parent(p, p->exit_signal))
1106 p->exit_state = state;
1107 write_unlock_irq(&tasklist_lock);
1109 if (state == EXIT_DEAD)
1113 infop = wo->wo_info;
1115 if ((status & 0x7f) == 0) {
1116 infop->cause = CLD_EXITED;
1117 infop->status = status >> 8;
1119 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1120 infop->status = status & 0x7f;
1129 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1132 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1133 return &p->exit_code;
1135 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1136 return &p->signal->group_exit_code;
1142 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1144 * @ptrace: is the wait for ptrace
1145 * @p: task to wait for
1147 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1150 * read_lock(&tasklist_lock), which is released if return value is
1151 * non-zero. Also, grabs and releases @p->sighand->siglock.
1154 * 0 if wait condition didn't exist and search for other wait conditions
1155 * should continue. Non-zero return, -errno on failure and @p's pid on
1156 * success, implies that tasklist_lock is released and wait condition
1157 * search should terminate.
1159 static int wait_task_stopped(struct wait_opts *wo,
1160 int ptrace, struct task_struct *p)
1162 struct waitid_info *infop;
1163 int exit_code, *p_code, why;
1164 uid_t uid = 0; /* unneeded, required by compiler */
1168 * Traditionally we see ptrace'd stopped tasks regardless of options.
1170 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1173 if (!task_stopped_code(p, ptrace))
1177 spin_lock_irq(&p->sighand->siglock);
1179 p_code = task_stopped_code(p, ptrace);
1180 if (unlikely(!p_code))
1183 exit_code = *p_code;
1187 if (!unlikely(wo->wo_flags & WNOWAIT))
1190 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1192 spin_unlock_irq(&p->sighand->siglock);
1197 * Now we are pretty sure this task is interesting.
1198 * Make sure it doesn't get reaped out from under us while we
1199 * give up the lock and then examine it below. We don't want to
1200 * keep holding onto the tasklist_lock while we call getrusage and
1201 * possibly take page faults for user memory.
1204 pid = task_pid_vnr(p);
1205 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1206 read_unlock(&tasklist_lock);
1207 sched_annotate_sleep();
1209 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1212 if (likely(!(wo->wo_flags & WNOWAIT)))
1213 wo->wo_stat = (exit_code << 8) | 0x7f;
1215 infop = wo->wo_info;
1218 infop->status = exit_code;
1226 * Handle do_wait work for one task in a live, non-stopped state.
1227 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1228 * the lock and this task is uninteresting. If we return nonzero, we have
1229 * released the lock and the system call should return.
1231 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1233 struct waitid_info *infop;
1237 if (!unlikely(wo->wo_flags & WCONTINUED))
1240 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1243 spin_lock_irq(&p->sighand->siglock);
1244 /* Re-check with the lock held. */
1245 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1246 spin_unlock_irq(&p->sighand->siglock);
1249 if (!unlikely(wo->wo_flags & WNOWAIT))
1250 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1251 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1252 spin_unlock_irq(&p->sighand->siglock);
1254 pid = task_pid_vnr(p);
1256 read_unlock(&tasklist_lock);
1257 sched_annotate_sleep();
1259 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1262 infop = wo->wo_info;
1264 wo->wo_stat = 0xffff;
1266 infop->cause = CLD_CONTINUED;
1269 infop->status = SIGCONT;
1275 * Consider @p for a wait by @parent.
1277 * -ECHILD should be in ->notask_error before the first call.
1278 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1279 * Returns zero if the search for a child should continue;
1280 * then ->notask_error is 0 if @p is an eligible child,
1283 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1284 struct task_struct *p)
1287 * We can race with wait_task_zombie() from another thread.
1288 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1289 * can't confuse the checks below.
1291 int exit_state = READ_ONCE(p->exit_state);
1294 if (unlikely(exit_state == EXIT_DEAD))
1297 ret = eligible_child(wo, ptrace, p);
1301 if (unlikely(exit_state == EXIT_TRACE)) {
1303 * ptrace == 0 means we are the natural parent. In this case
1304 * we should clear notask_error, debugger will notify us.
1306 if (likely(!ptrace))
1307 wo->notask_error = 0;
1311 if (likely(!ptrace) && unlikely(p->ptrace)) {
1313 * If it is traced by its real parent's group, just pretend
1314 * the caller is ptrace_do_wait() and reap this child if it
1317 * This also hides group stop state from real parent; otherwise
1318 * a single stop can be reported twice as group and ptrace stop.
1319 * If a ptracer wants to distinguish these two events for its
1320 * own children it should create a separate process which takes
1321 * the role of real parent.
1323 if (!ptrace_reparented(p))
1328 if (exit_state == EXIT_ZOMBIE) {
1329 /* we don't reap group leaders with subthreads */
1330 if (!delay_group_leader(p)) {
1332 * A zombie ptracee is only visible to its ptracer.
1333 * Notification and reaping will be cascaded to the
1334 * real parent when the ptracer detaches.
1336 if (unlikely(ptrace) || likely(!p->ptrace))
1337 return wait_task_zombie(wo, p);
1341 * Allow access to stopped/continued state via zombie by
1342 * falling through. Clearing of notask_error is complex.
1346 * If WEXITED is set, notask_error should naturally be
1347 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1348 * so, if there are live subthreads, there are events to
1349 * wait for. If all subthreads are dead, it's still safe
1350 * to clear - this function will be called again in finite
1351 * amount time once all the subthreads are released and
1352 * will then return without clearing.
1356 * Stopped state is per-task and thus can't change once the
1357 * target task dies. Only continued and exited can happen.
1358 * Clear notask_error if WCONTINUED | WEXITED.
1360 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1361 wo->notask_error = 0;
1364 * @p is alive and it's gonna stop, continue or exit, so
1365 * there always is something to wait for.
1367 wo->notask_error = 0;
1371 * Wait for stopped. Depending on @ptrace, different stopped state
1372 * is used and the two don't interact with each other.
1374 ret = wait_task_stopped(wo, ptrace, p);
1379 * Wait for continued. There's only one continued state and the
1380 * ptracer can consume it which can confuse the real parent. Don't
1381 * use WCONTINUED from ptracer. You don't need or want it.
1383 return wait_task_continued(wo, p);
1387 * Do the work of do_wait() for one thread in the group, @tsk.
1389 * -ECHILD should be in ->notask_error before the first call.
1390 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1391 * Returns zero if the search for a child should continue; then
1392 * ->notask_error is 0 if there were any eligible children,
1395 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1397 struct task_struct *p;
1399 list_for_each_entry(p, &tsk->children, sibling) {
1400 int ret = wait_consider_task(wo, 0, p);
1409 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1411 struct task_struct *p;
1413 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1414 int ret = wait_consider_task(wo, 1, p);
1423 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1424 int sync, void *key)
1426 struct wait_opts *wo = container_of(wait, struct wait_opts,
1428 struct task_struct *p = key;
1430 if (!eligible_pid(wo, p))
1433 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1436 return default_wake_function(wait, mode, sync, key);
1439 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1441 __wake_up_sync_key(&parent->signal->wait_chldexit,
1442 TASK_INTERRUPTIBLE, p);
1445 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1446 struct task_struct *target)
1448 struct task_struct *parent =
1449 !ptrace ? target->real_parent : target->parent;
1451 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1452 same_thread_group(current, parent));
1456 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1457 * and tracee lists to find the target task.
1459 static int do_wait_pid(struct wait_opts *wo)
1462 struct task_struct *target;
1466 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1467 if (target && is_effectively_child(wo, ptrace, target)) {
1468 retval = wait_consider_task(wo, ptrace, target);
1474 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1475 if (target && target->ptrace &&
1476 is_effectively_child(wo, ptrace, target)) {
1477 retval = wait_consider_task(wo, ptrace, target);
1485 static long do_wait(struct wait_opts *wo)
1489 trace_sched_process_wait(wo->wo_pid);
1491 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1492 wo->child_wait.private = current;
1493 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1496 * If there is nothing that can match our criteria, just get out.
1497 * We will clear ->notask_error to zero if we see any child that
1498 * might later match our criteria, even if we are not able to reap
1501 wo->notask_error = -ECHILD;
1502 if ((wo->wo_type < PIDTYPE_MAX) &&
1503 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1506 set_current_state(TASK_INTERRUPTIBLE);
1507 read_lock(&tasklist_lock);
1509 if (wo->wo_type == PIDTYPE_PID) {
1510 retval = do_wait_pid(wo);
1514 struct task_struct *tsk = current;
1517 retval = do_wait_thread(wo, tsk);
1521 retval = ptrace_do_wait(wo, tsk);
1525 if (wo->wo_flags & __WNOTHREAD)
1527 } while_each_thread(current, tsk);
1529 read_unlock(&tasklist_lock);
1532 retval = wo->notask_error;
1533 if (!retval && !(wo->wo_flags & WNOHANG)) {
1534 retval = -ERESTARTSYS;
1535 if (!signal_pending(current)) {
1541 __set_current_state(TASK_RUNNING);
1542 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1546 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1547 int options, struct rusage *ru)
1549 struct wait_opts wo;
1550 struct pid *pid = NULL;
1553 unsigned int f_flags = 0;
1555 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1556 __WNOTHREAD|__WCLONE|__WALL))
1558 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1570 pid = find_get_pid(upid);
1573 type = PIDTYPE_PGID;
1578 pid = find_get_pid(upid);
1580 pid = get_task_pid(current, PIDTYPE_PGID);
1587 pid = pidfd_get_pid(upid, &f_flags);
1589 return PTR_ERR(pid);
1598 wo.wo_flags = options;
1601 if (f_flags & O_NONBLOCK)
1602 wo.wo_flags |= WNOHANG;
1605 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1612 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1613 infop, int, options, struct rusage __user *, ru)
1616 struct waitid_info info = {.status = 0};
1617 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1623 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1629 if (!user_write_access_begin(infop, sizeof(*infop)))
1632 unsafe_put_user(signo, &infop->si_signo, Efault);
1633 unsafe_put_user(0, &infop->si_errno, Efault);
1634 unsafe_put_user(info.cause, &infop->si_code, Efault);
1635 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1636 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1637 unsafe_put_user(info.status, &infop->si_status, Efault);
1638 user_write_access_end();
1641 user_write_access_end();
1645 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1648 struct wait_opts wo;
1649 struct pid *pid = NULL;
1653 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1654 __WNOTHREAD|__WCLONE|__WALL))
1657 /* -INT_MIN is not defined */
1658 if (upid == INT_MIN)
1663 else if (upid < 0) {
1664 type = PIDTYPE_PGID;
1665 pid = find_get_pid(-upid);
1666 } else if (upid == 0) {
1667 type = PIDTYPE_PGID;
1668 pid = get_task_pid(current, PIDTYPE_PGID);
1669 } else /* upid > 0 */ {
1671 pid = find_get_pid(upid);
1676 wo.wo_flags = options | WEXITED;
1682 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1688 int kernel_wait(pid_t pid, int *stat)
1690 struct wait_opts wo = {
1691 .wo_type = PIDTYPE_PID,
1692 .wo_pid = find_get_pid(pid),
1693 .wo_flags = WEXITED,
1698 if (ret > 0 && wo.wo_stat)
1704 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1705 int, options, struct rusage __user *, ru)
1708 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1711 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1717 #ifdef __ARCH_WANT_SYS_WAITPID
1720 * sys_waitpid() remains for compatibility. waitpid() should be
1721 * implemented by calling sys_wait4() from libc.a.
1723 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1725 return kernel_wait4(pid, stat_addr, options, NULL);
1730 #ifdef CONFIG_COMPAT
1731 COMPAT_SYSCALL_DEFINE4(wait4,
1733 compat_uint_t __user *, stat_addr,
1735 struct compat_rusage __user *, ru)
1738 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1740 if (ru && put_compat_rusage(&r, ru))
1746 COMPAT_SYSCALL_DEFINE5(waitid,
1747 int, which, compat_pid_t, pid,
1748 struct compat_siginfo __user *, infop, int, options,
1749 struct compat_rusage __user *, uru)
1752 struct waitid_info info = {.status = 0};
1753 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1759 /* kernel_waitid() overwrites everything in ru */
1760 if (COMPAT_USE_64BIT_TIME)
1761 err = copy_to_user(uru, &ru, sizeof(ru));
1763 err = put_compat_rusage(&ru, uru);
1772 if (!user_write_access_begin(infop, sizeof(*infop)))
1775 unsafe_put_user(signo, &infop->si_signo, Efault);
1776 unsafe_put_user(0, &infop->si_errno, Efault);
1777 unsafe_put_user(info.cause, &infop->si_code, Efault);
1778 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1779 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1780 unsafe_put_user(info.status, &infop->si_status, Efault);
1781 user_write_access_end();
1784 user_write_access_end();
1790 * thread_group_exited - check that a thread group has exited
1791 * @pid: tgid of thread group to be checked.
1793 * Test if the thread group represented by tgid has exited (all
1794 * threads are zombies, dead or completely gone).
1796 * Return: true if the thread group has exited. false otherwise.
1798 bool thread_group_exited(struct pid *pid)
1800 struct task_struct *task;
1804 task = pid_task(pid, PIDTYPE_PID);
1806 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1811 EXPORT_SYMBOL(thread_group_exited);
1813 __weak void abort(void)
1817 /* if that doesn't kill us, halt */
1818 panic("Oops failed to kill thread");
1820 EXPORT_SYMBOL(abort);